Every CIO I know has had some version of this conversation: their CEO comes back from a golf trip with their buddy, or a conference with peers, and is told AI is about to automate everything at their company, from HR to marketing and finance. No humans in the loop, just AI. The CEO then calls an all-hands Monday morning, and the CIO is suddenly on the hook to make it all happen.

The instinct for CEOs to chase unsubstantiated claims is understandable since they’re responding to competitive pressure. But that leaves CIOs responsible to close the gap between ambition and reality. Making AI work in an organization with decades of accumulated process, permission frameworks, and cultural inertia is very different from deploying it in a demo.

The best response isn’t to push back on the ambition, but redirect it. Translate the CEOs vision into an honest map of what has to happen for the organization to get there, including the infrastructure, governance, and training. That helps to convert the kneejerk compulsion to move faster into a concrete plan that leadership can get behind.

Here’s what CIOs should actually be focused on to get where their CEOs want them to go, regardless of what’s discussed on the links.

1. Start where AI can build its own credibility

The hype machine wants you to climb Everest on day one. Instead, identify the repetitive tasks where AI can prove itself on familiar ground — the workflows your team already knows well, where results are easy to verify and the bar for trust is attainable.

The goal is the Eureka moment when a skeptic on your team sees a real result and becomes a believer. Those moments compound. When someone has seen AI make their work easier in a context they understand, they’re more likely to help you move things forward. You can’t force that change, but you can engineer the conditions for it.

2. Models will commoditize. Context will not.

Every few months, a new model claims to be smarter, faster, and cheaper than the last one. Don’t be distracted by that race. The lasting advantage in enterprise AI doesn’t just come from which model you’re running, it’s in the quality, governance, and semantic clarity of the data feeding it. Enterprises that invest in consistent business definitions, well-structured data, and clear lineage will outperform those that don’t, regardless of which model is in fashion. Context is your competitive moat. Focus on building that.

3. Nail down the permissions

In a world of dashboards, you know exactly what data will appear on a given page, so you can set permissions in advance for who can access it. In an AI world, the system can generate outputs that were never pre-designed. So how do you determine who has the right to see a result that was never anticipated?

Before deploying any agent that acts on someone’s behalf, such as filing a request, surfacing payroll data, or populating a record, first determine whether your existing permissions and access control frameworks can handle outputs that were never planned for. Most can’t. This is a prerequisite of what your CEO is asking for: the unglamorous infrastructure work that determines whether your AI is trustworthy in production. It needs to happen before you scale, not after.

4. Build an editing culture, not a writing one

For decades, engineers, analysts, and operations teams have been trained to write code, build reports, and define new processes. AI upends that. The skill now is editing — auditing what the system produces, catching what it got wrong, and knowing where to push back.

The truth is most people aren’t naturally good at editing because they’ve never had to be. That’s a skills gap that needs to be closed early on. Invest in helping engineers, analysts, and managers develop the judgment to evaluate AI outputs, not just generate them. Editing must become a core enterprise competency.

5. Measure behavior change, not tool adoption

Login data is a vanity metric. If your engineers are accessing AI coding tools but aren’t changing how they build, you haven’t adopted anything. The metric that makes more sense is productivity output. In agile terms, a team that completes 20 story points per sprint should hit about 28 with AI, not because the tools are magic, but because the repetitive work gets faster. If you’re not seeing that, you’re measuring the wrong thing. Pay attention to output, not usage metrics.

6. Reframe your organization’s relationship with failure

The instinct to de-risk everything made sense when software deployments were expensive and slow to reverse. AI works differently. The outputs are probabilistic, the iteration cycles are fast, and being overly cautious can cost valuable time. CIOs need to give teams permission to experiment in ways that feel uncomfortable by traditional enterprise standards, all while building the feedback loops that make fast failure safe. That culture shift has to be modeled from the top.

FOMO isn’t going away

CEOs will keep getting pulled into cycles of urgency and FOMO, and that pressure will keep landing on CIOs. The organizations that make real progress will be the ones that redirect that energy into infrastructure that makes AI trustworthy, measurement systems that show what’s working, and cultural changes that make adoption stick. That’s the agenda that’ll move your organization forward.

The IT department at Unum Group had a product management structure and worked in an agile delivery model.

This operating model gave IT teams and the company wins by rapidly delivering what they call “investment capabilities” that were aligned to the business.

But Shelia Anderson, who became executive vice president and chief information and digital officer in May 2025, saw room for improvement. She wanted to fine-tune her department’s operating structure to ensure investments deliver returns.

“There wasn’t a great correlation between those investments and that value recognition. So the part of it that needed some work was achieving value recognition in the business and making sure there was accountability for that,” Anderson says.

She also wanted accountability for improving time-to-value for investments, whether that value stemmed from productivity gains, improving customer experiences, or some other objective.

To do that, Anderson adopted a value stream model to analyze and optimize the end-to-end experience across each value chain within Unum Group, a provider of workplace benefits and services, including disability insurance, life insurance, and supplemental health products.

With that model shift, the company’s product-based approach became a business-owned one. Each value stream is now owned by a business leader, with the product management team associated with that value stream now reporting to that owner.

The core IT team assigned to each value stream includes a customer experience professional, a data lead, and an architecture lead. The team uses agile practices to deliver products along with product improvements, as products are part of the value stream.

“In insurance we have a lot of processes; products are within the larger processes,” Anderson explains, “and you have multiple processes that sit within a journey, so most of our products are within a value stream.”

Furthermore, the value stream model facilitates IT and business function collaboration to “make decisions around what we’re solving for and what are we going to deliver in this round of iteration,” she says. It also enables Unum Group to deliver end-to-end process improvements and change management as part of the deliverables, and measure the value delivered, she adds.

“The value stream concept is truly wrapping those all together,” Anderson says.

Rethinking the IT operating model

Anderson’s overhaul of the IT strategy at Unum Group showcases how CIOs are rethinking the IT operating model.

Many IT leaders are moving from traditional silos — security, development, support, etc. — to groupings designed around products, value streams, journeys, and customer lifecycles.

And they are doing so for several reasons, says Amar Aswatha, senior vice president for global business engineering at consulting and services firm CGI. To start, they have found that IT isn’t getting the business what it needs when it needs it. They have also found that IT’s work costs too much — “efficiency is slow as is productivity,” Aswatha says. And they are coming to realize that IT as traditionally configured can’t keep up with the pace of technology change and innovation.

As a result, CIOs are finding that a traditional structure focused on outputs (the delivery of a project, for example) instead of outcomes (for example, a specific, measurable improvement in business productivity) won’t get the business where it wants to go.

“So today CIOs are thinking about how to build a level of adaptability and agility in their operations model, and they’re thinking about how to build an organization that is continuously learning what’s working, where are the bottlenecks and points of frictions, and how they can get earlier signals on what’s not working so they can make adaptive changes,” says Fiona Mark, principal analyst at Forrester Research.

Ken Spangler, an instructor at Carnegie Mellon University’s CIDO Program, sees a trend toward organizing IT operations in a hybrid centralized-federated structure organized around products, domains, or capabilities. Here, there are centralized platforms, but enablement of those platforms is federated. For example, IT creates and maintains AI as a platform (the centralized component) but has IT working in business-facing teams to enable the various uses of the AI platform to deliver business value (the federated part).

The centralized IT work includes engineering and security teams, Spangler says. Product teams, which include product owners and business roles, support the federated work. CIOs ensure there’s a governance structure to manage both sides.

“In the AI era, it’s about product, platform, and governance: product for speed, platform for scale, and governance for control and risk management,” adds Spangler, who formerly served as executive vice president and CIO of FedEx Global Operations Technology.

Making the shift

For Anderson, Unum’s shift to a value stream model has required “a reimagining for roles” within the IT department as well as the expectations for those roles, she says. It also has IT teams thinking about the next evolution of agile and how they’ll use it to improve the work they’re aiming to do.

Sharing possibilities give Anderson the opportunity to leverage that centralized-federated approach within this value stream model as well. “There are some components of a value stream that could be shared,” she notes, pointing to the company’s data layer and integration layers that also exist to enable and support the products that are part of the value stream.

At Unum, teams are no longer static, with some IT workers assigned to multiple value steam core teams, Anderson says. To support IT professionals working in this new value stream model, Anderson will likely also adopt a chapter model, where IT employees are organized by discipline. That way chapter members who work on different value stream teams can come together to develop skills, foster expertise, define standards, and advance their careers.

So far, Unum’s shift to a value stream model has been incremental, with the first iteration having been completed in the first quarter of 2026. Still, Anderson is confident that the move will yield benefits.

For example, having a single value stream owner creates a higher level of accountability for ensuring investments deliver value.

“They know the north star of the value stream,” she says, adding that this empowers value stream owners to make quicker, better decisions around starting, continuing, or stopping investments. The model works well with a persistent funding approach as well as using metrics and score cards for measuring benefits and ROI. All that in turn helps teams “have a clear understanding of the value stream and what’s expected.”

“It’s truly shifting culture, so there is a clear structure around what result do we want, what is the process change, what’s the change needed with people, and then what’s the technology that’s needed. It’s getting the right people in the room to make those decisions,” Anderson says. “It’s truly business and technology at the table doing that design.”

Getting to the big picture

After starting as CIO of Tungsten Automation in February 2025, Shelley Seewald restructured her IT department into three components: business operations, enterprise IT delivery, and IT operations. The IT department had had a traditional operating model structured around technology, with a Salesforce team, a financial systems team, and the like.

Seewald felt a shakeup in how IT operates would move the organization away from being order-takers, which she says leads to an “inefficient and ineffective use of technology.” Her new structure breaks down silos and allows IT to “see the bigger picture, to see the ecosystem, to connect dots, and to spot opportunities,” she says.

Now each IT delivery team is aligned to a commercial organization (sales, marketing, customer support, etc.) or a back-office function (finance, legal, HR, etc.).

“The teams meet with them weekly, prioritize work, learn the business,” Seewald says. “This allows us [the IT department] to be a really a good technology partner. We’re there to understand the business first and then we offer them AI or technology solutions to help them reach their goals.”

IT operations is its own group comprising networking, help desk, and the like, Seewald adds. But even IT operations is expected to know the business side of the house. “They don’t align with a business per se, but we have them meet with IT delivery teams so they know what’s happening with the business as well and so they know about product introductions, new offices, and such.”

Challenges to tackle

Many CIOs have yet to move their IT operations from a conventional structure to one focused on products or value streams, says Rob Holbrook, principal of technology strategy and architecture with professional services firm Slalom.

The Global Tech Agenda 2026 from consultancy McKinsey & Co. reports that only about “one in ten top-performing companies have fully adopted product and platform models across all teams, which is more than four times that of other organizations. And nearly half of these companies indicate that at least half of their teams now operate this way.”

Such figures are not surprising, given the challenges that come with shifting how an organization operates.

For a shift to work, Holbrook says some CIOs and their teams must cultivate a true product mindset where IT leaders and workers have a clear understanding of the product IT delivery model.

CIOs also need to put in place a strong governance program to guide product teams, the business units, and the IT department in how to successfully work under such a model, Holbrook says. “They have to learn how to navigate it, and how to get needs prioritized,” he adds.

And CIOs should ensure that the focus on products and business outcomes doesn’t allow back-office needs to slip through the cracks, lest they end up with shadow IT filling those gaps, Holbrook says.

Powering growth

Julie Averill says moving to a modern IT operating model can produce significant value for an organization.

Averill modernized the IT operating model at Lululemon while working as executive vice president and global CIO from 2017 to 2025. She transitioned the department to product model mode, a shift that moved IT workers away from delivering initiatives to teams that owned products and the outcomes they were meant to generate.

“The business, management, and the product teams were all aligned along a mission and an outcome,” Averill explains. “The goal was to keep these teams together to work on outcomes but also have enough elasticity for team members to move to other products as business objectives changed.”

In this structure, Averill had centralized platform teams supporting shared infrastructure and capabilities, such as infrastructure, networking, and security. These teams, she notes, “became internal service teams.”

Averill, nowCEO of Gold Thread LLC and author of the book Chief Impact Officer, says changing the way IT operates “was an exercise in leadership.” It required convincing executive colleagues and business teams that the collaboration contributions they’d have to make and the new ways they’d have to fund product work would produce better results for the company. And it required hiring “technology-minded businesspeople and business-minded technologists who can understand and speak to the business but also can talk tech.”

Averill also says she needed to create professional communities, such as an engineering community, to support skills, standards, and a positive career experience for IT workers.

But the work was worth it, she says, crediting the changes she made while CIO with helping Lululemon grow from $2 billion to more than $10 billion in annual revenue over eight years.

IDC에 따르면 지난해 말 기준 2,800만 개 이상의 AI 에이전트가 배포됐으며, 2029년에는 10억 개 이상이 실제 운영 환경에서 활용되면서 하루 2,170억 건의 작업을 수행할 것으로 전망된다.

매출 46억 달러(약 6조 7,500억 원) 규모의 글로벌 신용평가 기업 트랜스유니온(TransUnion)의 최고 기술·데이터·분석 책임자 벤캇 아찬타는 “AI 에이전트 PoC(Proof of Concept)을 구축하는 것은 쉽다”라며 “하지만 이를 통제하고, 보안을 확보하며, 확장하는 것은 완전히 다른 차원의 과제”라고 말했다. 특히 금융 서비스와 헬스케어처럼 규제가 엄격한 산업일수록 이러한 어려움은 더욱 크다고 설명했다.

이 문제를 해결하기 위해 트랜스유니온은 지난 3년간 에이전틱 AI 플랫폼 ‘원트루(OneTru)’를 구축했다. 목표는 기존의 규칙 기반 전문가 시스템처럼 신뢰성과 예측 가능성을 확보하면서도, 생성형 AI처럼 유연하고 챗봇처럼 쉽게 사용할 수 있는 환경을 만드는 것이었다.

핵심은 두 접근 방식의 장점을 결합하는 데 있었다. 설명 가능성과 안정성이 중요한 핵심 업무는 전통적인 시스템이 담당하고, 생성형 AI는 특화된 작업에 한해 제한적으로 적용하는 방식이다. 이를 구현할 인프라가 시장에 존재하지 않았던 만큼, 트랜스유니온은 약 1억 4,500만 달러(약 2,100억 원)를 투자해 자체 구축에 나섰다.

검증되지 않은 기술에 대한 대규모 투자였지만, 이미 약 2억 달러(약 2,800억 원)의 비용 절감 효과를 거뒀다. 더 나아가 해당 플랫폼을 기반으로 고객용 솔루션까지 개발했다.

대표적으로 올해 3월, 트랜스유니온은 구글 제미나이 모델을 기반으로 원트루 플랫폼에서 구축한 ‘AI 애널리틱스 오케스트레이터 에이전트’를 공개했다. 이 에이전트는 내부 분석 효율을 높이는 데 활용되고 있으며, 고객 역시 데이터 과학자 없이 고급 데이터 분석을 수행할 수 있도록 지원한다.

아찬타는 “많은 고객이 트랜스유니온 데이터를 사용하면서도 다른 솔루션이나 플랫폼은 활용하지 않는다”라며 “이번 오케스트레이터 에이전트는 데이터 활용 가치를 높이고 새로운 수익원을 창출할 가능성이 있다”고 말했다.

현재 추가적인 에이전트도 개발 중이다. 아찬타는 “에이전트의 성능을 좌우하는 핵심은 오케스트레이션, 거버넌스, 보안 계층”이라며 “단순히 에이전트를 만드는 것은 며칠이면 가능하지만, 이를 안정적으로 운영하는 기반과 통제 장치가 진짜 경쟁력”이라고 강조했다. 이어 “플랫폼 위의 에이전트는 모든 가드레일과 기반을 활용하도록 설계돼 있으며, 이것이 우리의 힘”이라고 덧붙였다.

AI 에이전트를 효과적으로 통제하기 위한 핵심 전략은 작업을 여러 계층으로 분리하고, 각 계층을 서로 다른 시스템에 할당하는 것이다. 각 시스템은 일정한 제약 조건 아래 동작하며, 이를 통해 개별 에이전트의 영향 범위를 제한하고 전체 시스템에 견제와 균형 구조를 만든다. 또한 위험도가 높은 작업은 생성형 AI 이전 기술에 맡겨 리스크를 낮춘다.

트랜스유니온의 경우 핵심 의사결정은 업그레이드된 전문가 시스템이 담당한다. 이 시스템은 명확하게 정의되고 감사 가능한 규칙에 따라 동작하며, 예측 가능하고 비용 효율적이며 지연 시간도 낮다. 새로운 상황이 발생하면 LLM이 이를 분석하고, 다른 에이전트가 이를 새로운 규칙으로 변환한 뒤 인간이 검토해 최종적으로 전문가 시스템에 반영한다. 이 외에도 시맨틱 계층을 이해하거나 인간과 상호작용하는 등 다양한 역할을 수행하는 에이전트가 존재한다.

아찬타는 “신경망 기반 추론 계층인 LLM에는 인간을 개입시키고, 논리와 머신러닝 기반의 상징적 추론 계층은 자동화한다”고 설명했다.

이처럼 각 에이전트가 제한된 데이터와 역할 내에서 엄격한 제약을 가지고 동작하면, 전체 시스템은 훨씬 더 통제 가능하고 신뢰성 높은 구조로 발전한다.

이는 하나의 장인이 모든 작업을 수행하는 공방보다, 여러 작업자가 각자 역할을 나눠 수행하는 생산 라인에 비유할 수 있다. 생산 라인은 더 빠르고 안정적으로 작업을 수행할 수 있지만, 현재 많은 기업은 여전히 AI 에이전트를 장인처럼 운영하고 있다. 이러한 방식은 창의적인 결과를 만들 수 있지만, 기업 환경에서는 항상 적합한 선택은 아니다.

툴레인대학교 교수이자 ACM AI 특별 관심 그룹 의장인 니콜라스 마테이는 에이전트 시스템 간 연결 지점에서 보안을 강화해야 한다고 조언했다.

그는 “시스템 간 연결 지점마다 보안을 확보해야 한다”라며 “예를 들어 에이전트가 이메일 서비스에 요청을 보내는 경우, 두 시스템 사이에 검증 단계(체크포인트)를 두는 것이 필요하다”고 말했다. 이어 “신뢰하기 어려운 에이전트와 기존 소프트웨어가 만나는 경계 지점이 바로 보안 통제를 집중해야 할 영역”이라고 강조했다.

에이전틱 AI를 위한 보안 기반 구축

자동화 솔루션 기업 지터빗(Jitterbit)이 올해 3월 공개한 설문조사에 따르면, 1,500명의 IT 리더들은 AI 도입 최종 결정에서 가장 중요한 요소로 ‘AI 책임성’을 꼽았다. 이는 보안, 감사 가능성, 추적성, 가드레일 등을 포함하는 개념으로, 구현 속도나 벤더 평판, 심지어 총소유비용(TCO)보다도 높은 우선순위를 차지했다. 또한 보안, 거버넌스, 데이터 프라이버시 리스크는 비용이나 통합 문제보다도 AI 프로젝트의 운영 전환을 가로막는 주요 요인으로 나타났다. 이러한 우려는 충분히 근거가 있다.

실제 올해 초 사이버 보안 기업 코드월(CodeWall) 연구진은 맥킨지의 신규 AI 플랫폼 ‘릴리(Lilli)’를 침해하는 데 성공했다. 연구진은 자체 AI 도구를 활용해 4,700만 건의 채팅 메시지, 72만 8,000개 파일, 38만 4,000개의 AI 어시스턴트, 9만 4,000개 워크스페이스, 21만 7,000건의 에이전트 메시지, 약 400만 개에 달하는 RAG 문서 조각, 그리고 95개의 시스템 프롬프트 및 AI 모델 설정 정보에 접근할 수 있었다고 밝혔다.

연구진은 “수십 년간 축적된 맥킨지의 독점 연구와 프레임워크, 방법론이 누구나 읽을 수 있는 데이터베이스에 그대로 노출돼 있었다”며 “기업의 핵심 지식 자산이 사실상 무방비 상태였다”고 지적했다.

문제의 원인은 단순했다. 200개가 넘는 공개 API 엔드포인트 가운데 22개가 인증 절차 없이 열려 있었던 것이다. 연구진은 단 2시간 만에 릴리의 전체 운영 데이터베이스에 읽기 및 쓰기 권한을 확보했다. 이후 맥킨지는 즉각 대응에 나서 인증되지 않은 엔드포인트를 차단하고 추가 보안 조치를 시행했다.

맥킨지는 공식 성명을 통해 “외부 포렌식 전문기관과 함께 진행한 조사 결과, 해당 연구자나 다른 비인가 제3자가 고객 데이터 또는 기밀 정보를 실제로 열람했다는 증거는 발견되지 않았다”고 밝혔다.

IDC는 이번 사건이 AI 시스템 보안 침해가 기업에 얼마나 치명적인 영향을 미칠 수 있는지를 보여주는 사례라고 분석했다.

IDC AI 리서치 부문 부사장 알레산드로 페릴리는 “대부분의 기업은 여전히 데이터 유출, 잘못된 출력, 브랜드 평판 훼손 등 기존 관점에서 AI 리스크를 바라보고 있다”라며 “물론 중요한 문제지만, 더 큰 위험은 AI 시스템에 의사결정 권한을 위임하는 데 있다”고 강조했다.

에이전틱 AI 플랫폼에 대한 접근 권한을 확보할 경우, 공격자는 단순히 비인가 정보를 열람하는 데 그치지 않고 기업의 행동 방식 자체를 은밀하게 바꿀 수 있다. 또한 릴리(Lilli)와 같은 엔터프라이즈급 에이전틱 AI 시스템을 보호하는 것은 전체 과제의 절반에 불과하다. 가트너에 따르면 69%의 조직이 직원들이 금지된 AI 도구를 사용하고 있다고 의심하고 있으며, 이로 인해 2030년까지 40%의 조직이 보안 또는 규정 준수 사고를 겪을 것으로 예상된다.

그러나 현재의 탐지 도구만으로는 AI 에이전트를 충분히 식별하기 어렵다고 가트너는 지적한다.

현재 수천 개의 AI 에이전트를 운영 중인 KPMG의 글로벌 AI 및 데이터 랩 총괄 스와미나단 찬드라세카란은 “지금 기업 내에서 얼마나 많은 에이전트가 실행되고 있는지 묻는다면 어디에서 확인할 수 있겠느냐”라며 “이들이 모두 온보딩돼 정체성을 부여받았는지, 적절한 인증 절차를 거쳤는지, 누가 관리하는지 확인할 수 있는 인프라는 아직 존재하지 않는다”고 말했다.

그는 이어 “관련 도구들이 이제 막 등장하고 있거나 기업들이 자체적으로 구축하는 단계”라며 “이러한 체계가 CIO에게 안정감을 제공하게 될 것”이라고 덧붙였다.

이미 개인 직원이 강력한 에이전틱 AI를 도입해 부정적인 결과를 초래한 사례도 공개되고 있다. 메타(Meta)의 얼라인먼트 디렉터 서머 위는 최근 오픈소스 에이전틱 AI 도구 ‘오픈클로(OpenClaw)’를 이메일 관리에 활용하기로 결정했다. 테스트 환경에서 정상적으로 작동한 이후 실제 업무에 적용했다.

메타의 서머 위는 지난 2월 “작업 전 확인을 하도록 설정했음에도, 순식간에 받은 편지함을 삭제하는 모습을 보며 크게 당황했다”라며 “휴대폰으로는 중단할 수 없어 폭탄을 해체하듯 맥 미니로 달려가야 했다”고 X를 통해 전했다.

과거에는 직원이 민감한 정보를 챗봇에 입력하거나 보고서를 작성하게 한 뒤 이를 복사해 사용하는 수준에 머물렀다. 그러나 챗봇이 완전한 에이전트형 시스템으로 발전하면서 이제 에이전트는 사용자 권한 범위 내에서 가능한 모든 작업을 수행할 수 있으며, 기업 시스템에 접근하는 것까지 가능해졌다.

EY의 디지털 및 신기술 부문 책임자 라케시 말호트라는 이러한 새로운 보안 리스크를 관리하기 위해 기업들이 기존의 역할 기반 및 신원 기반 통제를 넘어 ‘의도 기반 통제’로 전환해야 한다고 강조했다.

그는 “에이전트가 시스템에 접근해 데이터를 변경할 권한이 있는지만 확인하는 것으로는 충분하지 않다”라며 “왜 그 변경을 수행하는지까지 확인할 수 있어야 한다”고 설명했다.

이어 “현재 관측 시스템은 에이전트의 행동 의도를 포착하지 못한다”라며 “신뢰는 의도에서 비롯되지만, 이를 측정할 수 있는 방법이 없는 상황”이라고 지적했다.

또 “만약 사람이 전체 코드베이스를 리팩토링하려 한다면 그 이유를 설명해야 한다”라며 “명확한 이유 없이 그런 작업을 진행해서는 안 된다. 사람의 경우 이를 판단할 방법이 있지만, 에이전트에는 아직 그런 체계가 없다”고 덧붙였다.

에이전틱 AI를 위한 시맨틱 데이터 기반 구축

트랜스유니온의 벤캇 아찬타는 자사의 원트루(OneTru) 플랫폼에서 ‘시맨틱 기반’의 중요성을 반복적으로 강조했다. 시맨틱 기반은 데이터가 무엇인지뿐 아니라 그 의미와 다른 데이터와의 관계까지 이해하도록 돕는 구조다. 가트너는 AI를 도입하는 기업이라면 시맨틱 레이어 구축이 이제 필수 과제라고 지적한다.

가트너는 “시맨틱 레이어는 정확도를 높이고 비용을 관리하며 AI 부채를 크게 줄이는 동시에, 멀티 에이전트 시스템을 정렬하고 비용이 큰 불일치를 사전에 차단할 수 있는 유일한 방법”이라고 설명했다.

또한 가트너는 2030년까지 범용 시맨틱 레이어가 데이터 플랫폼, 사이버 보안과 함께 핵심 인프라로 자리 잡을 것으로 전망했다. KPMG의 스와미나단 찬드라세카란은 “에이전트가 데이터를 활용해 의미 있는 작업을 수행하려면 맥락이 필수적”이라며 “그 안에 기업의 지식이 담겨 있다”고 말했다.

그는 이어 “이것이 기업의 새로운 지식재산(IP)”이라며 “맥락이 곧 새로운 경쟁력”이라고 강조했다.

미 법률 회사 굴스턴앤스토어스(Goulston & Storrs)의 CIO 존 아르스노는 견고한 데이터 기반 구축이 벤더 종속을 피하는 방법이기도 하다고 설명했다.

그는 “워크플로 자동화나 에이전틱 업무 지원을 위해 특정 솔루션에 데이터를 옮겨 넣으면, 이후 빠져나오기 매우 어려워진다”라며 “반면 데이터 중심 접근 방식을 취하면 시장 변화에 따라 다른 솔루션으로 유연하게 이동할 수 있다”고 말했다.

이 로펌은 고객 관련 업무 데이터를 법률 특화 문서 관리 시스템인 넷도큐먼츠(NetDocuments)로 이전했으며, 기타 데이터는 엔테그라타(Entegrata)의 법률 데이터 레이크하우스에 저장하고 있다.

아르스노는 “궁극적으로 모든 애플리케이션이 이 데이터 레이크를 중심으로 연결되도록 하는 것이 목표”라며 “이렇게 되면 회사의 모든 데이터가 두 개의 환경에 통합되고, 그 위에 어떤 AI 도구든 자유롭게 적용할 수 있다”고 설명했다.

이어 “데이터 흐름 관리도 훨씬 쉬워지고, 향후 등장할 AI 기술에도 빠르게 대응할 수 있다”라며 “생성형 AI든, 에이전틱 AI든, 혹은 앤트로픽 기반 기술이든 변화 속도가 너무 빨라 따라잡기 어렵다. 실제로 6개월마다 상황이 달라지고 있다”고 덧붙였다.

에이전트 오케스트레이션

보안 가드레일을 구축하고 활용 가능한 데이터 레이어를 마련한 이후, 에이전트 인프라 퍼즐의 마지막 단계는 ‘오케스트레이션’이다. 에이전틱 AI 시스템은 에이전트 간 상호작용은 물론, 인간 사용자와의 협업, 다양한 데이터 소스 및 도구와의 연동이 필요하다. 이는 매우 복잡한 과제로, 기술은 빠르게 발전하고 있지만 아직 초기 단계에 머물러 있다. MCP(Model Context Protocol)는 이러한 오케스트레이션 문제를 해결하기 위한 핵심 요소 중 하나로 꼽히며, AI 벤더들도 이 분야에서 협력적인 태도를 보이고 있다.

디지털 전환 기업 글로번트(Globant)의 디지털 혁신 수석 부사장이자 기술 담당 부사장인 아구스틴 우에르타는 “소셜 네트워크 초기, 페이스북과 트위터가 상호작용 표준을 논의할 때는 경쟁사의 프로토콜을 채택하려는 기업이 없었다”라며 “하지만 지금은 모두가 MCP를 중심으로 표준을 발전시키고 있다”고 말했다.

그러나 에이전트 통합 문제가 완전히 해결된 것은 아니다. 800명 이상의 IT 의사결정자와 개발자를 대상으로 한 도커(Docker) 설문조사에 따르면, 여러 구성 요소를 조율하는 운영 복잡성이 에이전트 구축의 가장 큰 과제로 나타났다.

구체적으로 응답자의 37%는 오케스트레이션 프레임워크가 운영 환경에 적용하기에는 아직 불안정하거나 미성숙하다고 답했으며, 30%는 복잡한 오케스트레이션 환경에서 테스트 및 가시성 부족을 문제로 지적했다.

또한 85%의 팀이 MCP를 인지하고 있음에도 불구하고, 실제 운영 환경 적용을 가로막는 보안, 구성, 관리 측면의 문제도 여전히 존재하는 것으로 나타났다. 이 외에도 기업이 해결해야 할 통합 과제는 적지 않다.

우에르타는 “아직 해결되지 않은 문제 중 하나는 모든 에이전트를 통합적으로 제어하고 상태를 파악할 수 있는 대시보드”라며 “오픈AI 기반 에이전트를 모니터링하는 도구와 세일즈포스 기반 에이전트를 관리하는 도구는 각각 존재하지만, 제어·감사·로깅을 위한 텔레메트리를 하나의 중앙 대시보드에서 통합 제공하는 솔루션은 없다”고 지적했다.

그는 이어 “단일 플랫폼에서 에이전트를 운영하거나 도입 초기 단계에서는 큰 문제가 아니지만, 에이전트 네트워크가 확장될수록 이러한 한계가 본격적으로 드러난다”고 설명했다. 실제로 글로번트는 자체적인 에이전트 AI 통합 대시보드를 개발 중이다.

한편 미국 전역에 고객을 둔 약 700명 규모의 로펌 브라운스타인 하얏트 파버 슈렉(Brownstein Hyatt Farber Schreck)은 제안서 생성 시스템 등 다양한 영역에 AI를 적용하고 있다.

이 회사의 CIO 앤드루 존슨은 “기존에는 고객 제안요청서(RFP)를 검토하고, 수기 메모나 회의 기록을 분석한 뒤 관련 자료를 정리하는 데 며칠이 걸렸다”라며 “이제는 모든 정보를 시스템에 입력해 핵심 기준을 추출하고 몇 분 만에 수준 높은 초안을 생성할 수 있다”고 말했다.

이 과정에는 여러 에이전트가 협력한다. 성공 기준이나 인력 요건을 추출하는 에이전트, 과거 사례와 교훈을 분석하는 에이전트, 가격 책정과 브랜드 기준을 담당하는 에이전트 등이 각각 역할을 수행한다. 존슨은 “각 에이전트는 독립적으로 동작하지만, 결과물이 다음 단계로 이어지도록 반드시 오케스트레이션이 필요하다”고 설명했다. 현재는 대부분 기존 시스템에 MCP 레이어가 없기 때문에 RAG 기반 구조를 활용하고 있다.

또한 작업에 따라 서로 다른 AI 모델이 사용되기도 하는데, 이 역시 추가적인 오케스트레이션 관리 요소로 작용한다.

비용 관리도 중요한 이슈다. AI 에이전트가 무한 피드백 루프에 빠질 경우 추론 비용이 급격히 증가할 수 있기 때문이다.

존슨은 “이러한 가능성을 인지하고 있으며, 아직 실제로 발생한 사례는 없지만 모니터링 체계를 구축해 임계치를 초과할 경우 즉각 대응하도록 하고 있다”고 말했다.

이처럼 다양한 대응 전략에도 불구하고, AI를 둘러싼 변화 속도는 기업이 경험한 그 어떤 기술보다 빠르다.

EY의 말호트라는 “25년간 기술 업계에 있었지만 지금과 같은 변화는 처음”이라며 “역사상 가장 빠르게 성장한 기업들이 최근 3~4년 사이에 등장했고, 기술 도입 속도 역시 전례가 없다”고 말했다. 이어 “불과 9~10개월 전까지만 해도 핵심이었던 기술이 이미 지나간 사례도 많다”고 덧붙였다.
dl-ciokorea@foundryco.com

IDC estima que a finales del año pasado había más de 28 millones de agentes de IA desplegados, y predice que en 2029 habrá más de 1.000 millones activos, ejecutando 217.000 millones de acciones al día.

Es fácil crear una prueba de concepto (POC) de un agente de IA, afirma Venkat Achanta, director de tecnología, datos y análisis de TransUnion, una empresa global de informes crediticios con unos ingresos de 4.600 millones de dólares. Pero gestionarlo, protegerlo y escalarlo supone todo un reto, especialmente para empresas de sectores altamente regulados, como los servicios financieros y la sanidad. Para abordar el problema, TransUnion ha dedicado los últimos tres años a desarrollar su plataforma de IA agentiva, OneTru. El objetivo era crear algo tan fiable y determinista como los antiguos sistemas basados en scripts y diseñados por expertos, pero tan flexible como la IA general, y tan fácil de interactuar como un chatbot.

El truco, sin embargo, consistía en combinar lo mejor de ambos mundos utilizando sistemas tradicionales para los procesos centrales, donde la explicabilidad y la fiabilidad son clave, e incorporando la funcionalidad de la IA general de forma limitada para las tareas para las que resultaba especialmente adecuada. Y dado que no se disponía de la infraestructura necesaria para ello, TransUnion construyó la suya propia, destinando 145 millones de dólares al proyecto. Fue una gran inversión en una tecnología sin probar, pero ya ha supuesto un ahorro de 200 millones de dólares. Más aún, una vez construida la plataforma, TransUnion la utilizó para crear soluciones orientadas al cliente.

En marzo de este año, por ejemplo, TransUnion lanzó su AI Analytics Orchestrator Agent, creado con la plataforma OneTru y basado en los modelos Gemini de Google. TransUnion ya utiliza este agente internamente para mejorar los análisis, y los clientes también pueden utilizarlo para realizar sofisticados análisis de datos sin necesidad de recurrir a científicos de datos.

Muchos clientes utilizan los datos de TransUnion, pero no utilizan otras soluciones ni plataformas, afirma Achanta. El nuevo agente de orquestación tiene el potencial de ayudar a los clientes a sacar más partido a los datos y de abrir nuevas fuentes de ingresos para la empresa. Y hay más agentes en desarrollo, afirma Achanta. La clave para que funcionen son las capas de orquestación, gobernanza y seguridad. Hacer que un agente haga algo es muy fácil para cualquiera, dice, y puede llevar solo unos días. La empresa también puede crear agentes rápidamente. “Pero yo tengo la base y las barreras de seguridad, y el agente que se encuentra en mi plataforma las utiliza todas. Eso es lo que nos da poder”, afirma.

El secreto para lograr que los agentes de IA se comporten es separar las capas de la tarea y asignar cada capa a un sistema diferente, cada uno de los cuales opera bajo un conjunto de restricciones. Este enfoque limita el daño que puede causar cualquier agente en particular, crea un sistema de controles y contrapesos, y restringe las actividades más arriesgadas a una tecnología de IA de generación previa.

Por ejemplo, en TransUnion, la toma de decisiones principal la lleva a cabo una versión actualizada de un sistema experto. Funciona bajo un conjunto de reglas bien definidas y auditables, y opera de forma predecible, rentable y con baja latencia. Cuando se encuentra con una situación que no ha visto antes, se utiliza un LLM para analizar el problema; a continuación, un agente diferente podría convertirlo en una nueva regla, y luego se podría recurrir a un humano para revisar los resultados antes de que la nueva regla se añada al sistema experto. Hay diferentes agentes que comprenden la capa semántica, interactúan con los humanos y realizan otras tareas. “Con la capa de razonamiento neuronal —el LLM— incorporamos a los humanos al proceso. Cuando se trata de una capa de razonamiento simbólico, que se basa en la lógica y el aprendizaje automático, dejamos que se automatice”, explica.

Así, cuando cada agente opera dentro de restricciones muy estrictas, solo con los datos limitados que necesita para esa tarea concreta, y está limitado a lo que puede hacer, todo el sistema se vuelve mucho más manejable y fiable. Es como la diferencia entre una cadena de montaje, donde varios trabajadores realizan cada uno una tarea única y distinta, en lugar de un taller donde un solo artesano lo hace todo. La cadena de montaje puede trabajar más rápido y de forma más fiable, pero hoy en día muchas empresas implementan sus agentes de IA como si fueran artesanos. Este último enfoque puede dar lugar a productos creativos y únicos, pero no siempre es lo que una empresa necesita.

Nicholas Mattei, presidente del grupo de interés especial de la ACM sobre IA y profesor de la Universidad de Tulane, sugiere que las empresas se centren en incorporar seguridad adicional en los puntos donde se conectan las diferentes partes del sistema de agentes. “Hay que asegurarse de que hay seguridad en las uniones”, afirma. Por ejemplo, si un agente envía solicitudes a un servicio de correo electrónico, hay que configurar un punto de control entre ambos. “En los huecos entre los agentes poco fiables y donde reside el software tradicional es donde se tienen que ubicar los procesos de seguridad”, relata.

Crear una base de seguridad para la IA agentiva

En una encuesta de Jitterbit realizada a 1.500 líderes de TI publicada en marzo, la responsabilidad de la IA —seguridad, auditabilidad, trazabilidad y medidas de protección— es el factor más importante a la hora de tomar la decisión final de compra de IA, por delante de la velocidad de implementación, la reputación del proveedor e incluso el coste total de propiedad. Los riesgos de seguridad, gobernanza y privacidad de los datos también fueron las principales cuestiones que impedían que las iniciativas de IA pasaran a producción, por delante de los costes y los retos de integración. Y tienen razón en estar preocupados.

A principios de este año, investigadores de la empresa de ciberseguridad CodeWall lograron vulnerar la nueva plataforma de IA de McKinsey, Lilli. Utilizando una herramienta de IA propia, los investigadores afirmaron que pudieron acceder a 47 millones de mensajes de chat, 728.000 archivos, 384.000 asistentes de IA, 94.000 espacios de trabajo, 217.000 mensajes de agentes, casi 4 millones de fragmentos de documentos RAG y 95 indicaciones del sistema y configuraciones de modelos de IA. “Se trata de décadas de investigación, marcos y metodologías propios de McKinsey: las joyas de la corona intelectual de la empresa almacenadas en una base de datos a la que cualquiera podía acceder”, escribieron los investigadores.

¿El motivo? De los más de 200 puntos finales de API expuestos públicamente, 22 no requerían autenticación. Los investigadores tardaron solo dos horas en obtener acceso completo de lectura y escritura a toda la base de datos de producción de Lilli. McKinsey respondió rápidamente a la alerta, corrigió los puntos finales sin autenticación y tomó otras medidas de seguridad. “Nuestra investigación, respaldada por una empresa forense externa líder, no identificó ninguna prueba de que este investigador o cualquier otro tercero no autorizado hubiera accedido a datos o información confidencial de los clientes”, afirmó la empresa en un comunicado.

IDC indica que el incidente pone de relieve lo peligrosa que puede ser la violación de un sistema de IA para una empresa. “La mayoría de las empresas siguen pensando en los riesgos de la IA en términos del pasado: fuga de datos, resultados erróneos y daño a la reputación de la marca”, explica Alessandro Perilli, vicepresidente de investigación en IA de IDC. “Esos son problemas graves, pero el mayor riesgo reside en delegar autoridad a los sistemas de IA”.

Al obtener acceso a una plataforma de IA agentiva, un atacante no solo puede ver algo que no debería, sino también cambiar de forma encubierta la forma de actuar de la empresa. Y proteger sistemas de IA agentiva a escala empresarial como Lilli es solo la mitad del reto. Según Gartner, el 69% de las organizaciones sospecha que sus empleados utilizan herramientas de IA prohibidas, y el 40% sufrirá incidentes de seguridad o de cumplimiento normativo para 2030 como consecuencia de ello.

Pero las herramientas de detección disponibles no están del todo preparadas para encontrar agentes de IA, indican desde Gartner. “Si te preguntara cuántos agentes se ejecutan en tu empresa en este momento, ¿dónde irías a buscarlo?”, pregunta Swaminathan Chandrasekaran, director global de IA y laboratorios de datos en KPMG, que ahora cuenta con varios miles de agentes de IA en producción. “¿Se han incorporado todos y tienen identidades? ¿Han pasado por un proceso de autenticación adecuado y quién está a cargo de ellos? Esa infraestructura no existe”.

Sin embargo, las herramientas están empezando a surgir, o las empresas están creando soluciones “hazlo tú mismo”, cuenta. “Eso es lo que va a dar tranquilidad a los directores de sistemas de información”. Ya estamos viendo ejemplos públicos de empleados individuales que implementan una potente IA agentiva con consecuencias negativas. Summer Yue, directora de alineación de Meta, decidió recientemente utilizar OpenClaw, una herramienta viral de IA agentiva de código abierto, para ayudarla a gestionar su bandeja de entrada. Después de que funcionara en una bandeja de entrada de prueba, la implementó de verdad.

“Nada te hace sentir más humilde que decirle a tu OpenClaw que confirme antes de actuar y ver cómo borra tu bandeja de entrada a toda velocidad”, escribió en X. “No pude detenerlo desde mi teléfono. Tuve que correr hacia mi Mac mini como si estuviera desactivando una bomba”. En el pasado, un empleado podía subir información confidencial a un chatbot o pedirle que redactara un informe que luego copiaría y pegaría, haciéndolo pasar por suyo. A medida que estos chatbots evolucionan hacia sistemas agenticos completos, los agentes tienen ahora la capacidad de hacer cualquier cosa para la que el usuario tenga privilegios, incluido el acceso a los sistemas corporativos.

Para gestionar este nuevo riesgo de seguridad, las empresas tendrán que pasar de controles basados en roles e identidades a otros basados en la intención, afirma Rakesh Malhotra, director de tecnologías digitales y emergentes en EY. No basta con preguntar si un agente tiene permiso para acceder a un sistema y realizar un cambio en un registro, afirma. Las empresas deben poder preguntar por qué se está realizando ese cambio. Ese es un gran reto en este momento. “La tecnología de observabilidad no capta la intención de por qué el agente ha hecho algo”, afirma. “Y eso es realmente importante de entender. La confianza se basa en la intención, y no hay forma de que ninguno de estos sistemas capte la intención”.

Si un empleado humano intentara refactorizar toda la base de código, se le pediría que diera una buena razón para hacerlo. “Y si estás refactorizando sin ninguna razón específica, quizá no deberías hacerlo”, dice Malhotra. “Con las personas, hay formas de juzgar esto. No sé cómo hacerlo con los agentes”.

Creación de una base de datos semántica para la IA agentiva

Achanta, de TransUnion, menciona repetidamente la base semántica de la plataforma OneTru de la empresa. Esa comprensión de la información ayuda a los sistemas a entender no solo qué son los datos, sino qué significan y cómo se relacionan con otros datos. Gartner afirma que desarrollar una capa semántica es ahora imprescindible para las empresas que implementan IA. “Es la única forma de mejorar la precisión, gestionar los costes, reducir sustancialmente la deuda de IA, alinear los sistemas multiagente y detener las costosas inconsistencias antes de que se extiendan”, dice.

Para 2030, las capas semánticas universales se considerarán infraestructura crítica, junto con las plataformas de datos y la ciberseguridad, predice Gartner. Y los agentes necesitan contexto para poder hacer algo significativo con los datos, afirma Chandrasekaran, de KPMG. Ahí es donde reside el conocimiento de una empresa. “Esa es tu nueva propiedad intelectual para la empresa. El contexto es la nueva muralla defensiva”.

Para John Arsneault, director de sistemas de información de Goulston & Storrs, crear una base de datos sólida es también una forma de evitar la dependencia de un proveedor. “Si compras productos y trasladas tus datos a ellos para automatizar flujos de trabajo o crear asistentes de trabajo para los agentes, te costará mucho salir de ahí. Pero si adoptas un enfoque centrado en los datos, al menos podrás pasar de uno a otro si se produce un cambio en el mercado”.

El bufete de abogados ha migrado sus productos de trabajo orientados al cliente a NetDocuments, un sistema de gestión de documentos enfocado específicamente al sector jurídico. Y el resto de los datos que recopila la empresa se almacenan en el ‘data lakehouse’ jurídico de Entegrata.

“Nuestro objetivo es que, con el tiempo, todas nuestras demás aplicaciones apunten a ese lago de datos. Entonces tendremos estos dos entornos donde residen todos los datos del bufete, lo que nos permitirá integrar cualquier herramienta de IA que utilicemos”, afirma.

También facilitará la gestión de los flujos de datos, añade, y permitirá al bufete adaptarse rápidamente a cualquier tecnología de IA que surja en el futuro. “Ya sea IA generativa, agéntica o de Anthropic, con el complemento legal de Cowork, es muy difícil mantenerse al día. Y cambia cada seis meses”.

Orquestación de agentes

La última pieza del rompecabezas de la infraestructura de agentes, tras establecer las medidas de seguridad y crear una capa de datos utilizable, es la orquestación. Los sistemas de IA de agentes requieren que los agentes se comuniquen entre sí y con los usuarios humanos, e interactúen con fuentes de datos y herramientas. Es un reto complicado, y esta tecnología se encuentra todavía en una fase muy incipiente, aunque avanza rápidamente. MCP es un ejemplo de ello, y es una pieza clave para resolver el rompecabezas de la orquestación. Los proveedores de IA se han mostrado muy dispuestos a cooperar en este ámbito.

“Cuando surgieron las redes sociales, y Facebook y Twitter debatían sobre un protocolo estándar para interactuar, nadie quería adoptar el protocolo de sus competidores”, afirma Agustín Huerta, vicepresidente sénior de innovación digital y vicepresidente de tecnología en Globant, una empresa de transformación digital. “Ahora todo el mundo está adoptando MCP y madurándolo como protocolo estándar”.

Pero eso no quiere decir que la integración de agentes se haya resuelto. Según una encuesta de Docker realizada a más de 800 responsables de la toma de decisiones de TI y desarrolladores, la complejidad operativa de orquestar múltiples componentes es el mayor desafío a la hora de crear agentes.

En concreto, el 37% de los encuestados afirma que los marcos de orquestación son demasiado frágiles o inmaduros para su uso en producción, y el 30% señala deficiencias en las pruebas y la visibilidad en orquestaciones complejas.

Además, aunque el 85% de los equipos están familiarizados con MCP, la mayoría afirma que existen importantes problemas de seguridad, configuración y gestionabilidad que impiden su implementación en producción. Y hay otros problemas de integración a los que las empresas deben hacer frente.

“Un problema aún por resolver es cómo conseguir un panel de control adecuado para gestionar todos estos agentes, para saber exactamente qué está pasando con cada uno de ellos”, afirma Huerta. “Hay un panel que permite supervisar los agentes creados con OpenAI y otro para los que residen en Salesforce, pero ninguno puede mostrar la telemetría en un panel centralizado para el control, la auditoría y el registro”.

Para las empresas que acaban de empezar a implementar agentes, o que se ciñen a una única plataforma, esto aún no supone un problema, añade, pero a medida que aprovechen una red más amplia de agentes, empezarán a experimentar estos retos. La propia Globant está creando su propio panel de control interno para la IA basada en agentes, por ejemplo.

Y en Brownstein Hyatt Farber Schreck, un bufete de abogados con 50 años de antigüedad, unos 700 empleados y clientes en todo Estados Unidos, hay varias áreas en las que se está implementando la IA, incluido un sistema generador de propuestas.

Normalmente, varias personas pueden tardar días en revisar la solicitud de propuesta de un cliente, examinar notas manuscritas o transcripciones de reuniones y recopilar otros materiales relevantes, afirma Andrew Johnson, director de sistemas de información del bufete. “Podemos introducir toda esa información en un ordenador y extraer los criterios clave para producir un primer borrador de calidad en cuestión de minutos”, afirma.

Se requieren múltiples agentes para las diferentes partes del proceso: uno para extraer los criterios de éxito o los requisitos de personal, otro para buscar precedentes y lecciones aprendidas, y otros para la fijación de precios y los estándares de marca. “Cada uno de esos agentes es autónomo y debe coordinarse para que los resultados de cada uno se incorporen al siguiente paso”, explica Johnson. En su mayor parte, eso significa un sistema RAG, ya que la mayoría de las plataformas heredadas que utiliza la empresa aún no han incorporado una capa MCP.

Dependiendo de la tarea, los agentes individuales pueden funcionar con diferentes modelos, lo que supone otra capa de coordinación que hay que gestionar. Luego está el control de costes. Si un agente de IA o un grupo de agentes entra en un bucle de retroalimentación infinito, los costes de inferencia pueden aumentar rápidamente. “Somos conscientes de la preocupación, aunque aún no la hemos visto materializarse”, afirma Johnson. “Por eso contamos con un sistema de supervisión. Si superamos los umbrales, reaccionamos”.

Independientemente de las estrategias o medidas para absorber los contratiempos, todo lo relacionado con la IA está cambiando más rápido que cualquier otra cosa que las empresas hayan visto. “Llevo 25 años en el sector tecnológico y nunca había visto nada igual”, señala Malhotra, de EY. “Las empresas de más rápido crecimiento de la historia se han creado todas en los últimos tres o cuatro años. El crecimiento en la adopción no tiene precedentes. Y hablo constantemente con clientes que están implementando tecnologías que eran muy relevantes hace nueve o diez meses, y todo el mundo ha pasado página”.

보안 기술 기업 넷스코프의 최고 디지털·정보 책임자 마이크 앤더슨은 IT 직원들에게 이례적인 과제를 부여했다. 각자의 역할을 반영한 ‘제미나이 젬(Gemini Gems)’ 디지털 트윈을 생성하고, 기술 문서 등 다양한 정보를 AI에 입력해 해당 역할의 업무와 필요 역량을 학습시키라는 것이다.

앤더슨은 이러한 AI 기반 디지털 트윈이 직원들의 업무 수행을 지원할 것으로 기대하고 있다. 간단한 질의만으로도 거의 실시간에 가까운 속도로 필요한 정보를 찾아볼 수 있도록 돕는다는 설명이다.

그는 “팀이 의지할 수 있는 전문가 역할의 젬(Gems)를 만들었다”라며 “각 직원이 일정 시간을 절약할 수 있도록 하는 것이 목표”라고 말했다.

이 같은 시도는 IT 부서의 효율성과 생산성을 높이기 위해 앤더슨이 추진 중인 다양한 워크플로우 및 프로세스 혁신 전략 중 하나다. 실제로 일부 성과도 나타나고 있다.

예를 들어 개발팀은 AI를 활용해 코드를 생성하고 있다. 직원들은 ‘바이브 코딩’을 통해 빠르게 초기 결과물을 만든 뒤 이를 반복 개선하는 방식으로 개발을 진행하며, 기존 제품 개발 일정에서 수개월을 단축하고 있다. 또한 앤더슨의 팀은 특정 요소, 특히 보안 통제가 AI 생성 코드에 항상 포함되도록 하는 ‘프리미티브’를 구축해 IT 인력의 업무 시간을 줄이고 있다.

앤더슨은 이러한 효율 향상에 대한 구체적인 투자 대비 효과(ROI)는 산출하지 않았지만, 결과적으로 더 적은 자원으로 더 많은 성과를 낼 수 있게 됐다고 강조했다. 그는 “예산을 유지한 상태에서도 이전보다 더 많은 기능과 결과물을 제공할 수 있다”고 말했다.

“현재 인력으로 더 많은 성과를”

CIO는 오랫동안 ‘적은 자원으로 더 많은 성과를 내야 한다’는 압박을 받아왔다. 그리고 그 압박은 지금 더욱 커지고 있다. 시장조사업체 가트너의 조사에 따르면 CIO의 57%는 생산성 향상, 52%는 비용 절감 요구에 직면해 있다.

동시에 CIO들은 기술, 특히 AI를 활용해 전사 워크플로우를 혁신하고 생산성과 효율성을 끌어올려야 하는 과제를 안고 있다. 이러한 변화는 IT 부서 내부에서도 동일하게 요구되고 있다.

네트워크 기업 익스트림 네트웍스의 최고 정보·고객 책임자 아니샤 바스와니는 “AI를 통해 IT 프로세스를 과감하게 재정의하고, 새로운 가능성을 탐색함으로써 가치를 창출해야 한다”고 말했다.

바스와니는 IT 워크플로우 혁신을 핵심 과제로 설정했다. 앤더슨과 마찬가지로 AI, 특히 클로드 코드를 활용해 코딩 속도를 높이고 있다. 또한 업무 방식을 재편해 직원들이 직접 코드를 작성하는 대신 프롬프트 설계, 결과 검토, 품질 관리에 집중하도록 전환했다.

아울러 다른 CIO들과 마찬가지로 헬프데스크 운영도 변화시키고 있다. AI와 자동화를 활용해 셀프서비스 비중을 확대하는 방식이다.

더 복잡한 워크플로우에서도 성과가 나타나고 있다. 바스와니는 AI를 활용해 테스트 전략을 생성하고 테스트를 자동화함으로써 IT의 QA 기능을 확장하고 있다. 그는 “수작업으로 수주가 걸리던 작업을 몇 분으로 줄일 가능성이 크다”고 설명했다. 또한 AI 활용을 통해 비용 증가 없이 처리 역량을 확대할 수 있을 것으로 보고 있다.

이와 함께 신규 제품 개발이나 기능 개선 과정에서 사용자 요구사항을 보다 효과적으로 수집하는 데 AI를 활용하는 방안도 모색하고 있다.

바스와니는 “IT가 비즈니스 파트너와 상호작용하는 방식을 재정의해 더 민첩하고 대응력 있게 만들고 싶다”라며 “이를 통해 더 자주 가치를 제공하고 고객 중심성을 강화할 수 있다”고 말했다. 이어 “목표는 현재 인력으로 더 많은 일을 해내는 것”이라며 “더 빠르게 혁신하고 더 많은 결과를 만들어내는 데 있다”고 강조했다.

변화 요구 커지는 IT, 핵심은 ‘워크플로우 재설계’

컨설팅 기업 웨스트 먼로의 디렉터 알렉스 와이어트는 IT 업무가 본질적으로 프로세스 중심이기 때문에 혁신 가능성이 큰 영역이라고 진단했다.

그는 “AI가 이 논의를 다시 촉발했다”라며 “현재 CIO들은 비용 절감 압박이 커지고 있고, 이사회는 ‘이 프로세스를 50% 더 효율적으로 만들어야 한다’고 요구하고 있다”고 말했다.

와이어트는 CIO들이 조직 전반과 마찬가지로 비교적 쉬운 과제부터 시작해 성과를 쌓고 역량을 확보한 뒤, 점차 난이도가 높은 영역으로 확장해야 한다고 조언했다. 그는 “워크플로우와 프로세스 최적화에는 여러 단계가 존재한다”고 설명했다.

초기 단계는 반복적인 업무를 AI로 자동화하고, 인력은 이를 감독하는 역할로 전환하는 것이다. 동시에 IT가 사용하는 도구와 기술에 이미 내장된 기능을 최대한 활용하는 것도 포함된다.

와이어트는 “이 단계가 가장 빠르게 성과를 낼 수 있는 영역이며, 투자 대비 효과도 가장 크다”라며 “이후 보다 고도화된 기회를 공략해야 한다”고 말했다.

다만 그는 AI가 유일한 해법은 아니라고 강조했다. “AI가 워크플로우와 프로세스 혁신을 부각시킨 것은 맞지만, 순수한 프로세스 개선 기회도 여전히 존재한다”며 린(Lean) 프로세스 설계를 예로 들었다.

이어 “비효율적인 프로세스를 그대로 자동화할 위험이 있기 때문에, 원하는 결과와 핵심성과지표(KPI)를 기준으로 업무를 어떻게 재구성할지 고민해야 한다”며 “단순히 도구를 늘린다고 효율이 높아지는 것이 아니라, 워크플로우와 업무 방식 자체를 재정의해야 한다”고 강조했다.

이 같은 조언은 오랜 기간 검증된 접근법이다. 와이어트는 “공격적인 워크플로우 및 프로세스 재설계를 통해 50% 이상의 성과 개선도 가능하다”라며 “이를 위해서는 업무 수행 방식 자체를 근본적으로 재검토하고 이를 뒷받침할 시스템을 구축해야 한다”고 말했다.

모든 CIO가 이러한 수준의 혁신을 추진할 수 있는 것은 아니지만, 그렇다고 시도를 미뤄서는 안 된다는 점도 짚었다. 그는 “기본적인 프로세스 재설계만으로도 10~20% 수준의 개선 효과를 얻을 수 있으며, 여기에 AI를 더하면 추가적인 성과를 기대할 수 있다”고 설명했다.

에베레스트 그룹의 파트너이자 CIO 리서치 및 자문을 총괄하는 로스 티스노프스키 역시 ‘변환’의 중요성을 강조했다. 그는 “워크플로우 재설계 없이 자동화만 진행할 경우 문제가 발생할 수 있다”고 지적했다.

티스노프스키는 코딩과 테스트 영역을 사례로 들었다. AI 도입으로 코딩 생산성은 70% 이상 향상된 반면, 테스트 효율은 약 30% 수준에 머물고 있다는 것이다. 이로 인해 워크플로우를 재구성하지 않은 조직에서는 코드 생성 속도가 테스트 처리 능력을 앞지르는 불균형이 발생한다.

그는 “많은 AI 프로젝트가 기대만큼의 가치를 창출하지 못하는 이유는 워크플로우를 함께 재설계하지 않았기 때문”이라고 분석했다.

이중 압박 속 IT 혁신, 핵심은 명확한 목표 설정

알렉스 와이어트는 CIO들이 IT 조직의 업무 방식을 혁신하는 과정에서 또 다른 도전에 직면한다고 지적했다.

우선 CIO와 IT 조직은 이미 기업 내 다른 부서들의 혁신 작업을 지원하는 데 상당한 자원을 투입하고 있다. 특히 매출 확대, 시장 점유율 상승, 고객 유지율 개선 등 직접적인 성과로 이어지는 영역이 우선순위를 차지하는 경우가 많다.

와이어트는 “IT는 독특한 위치에 있다. 전사 차원의 혁신을 추진하는 동시에 IT 내부 혁신도 요구받고 있다”라며 “이중의 압박을 받고 있는 셈”이라고 설명했다.

이로 인해 CIO들은 IT 워크플로우를 재설계하는 데 필요한 자원과 역량을 충분히 배분하기 어려운 상황에 놓이게 된다.

기존에 자리 잡은 워크플로우를 바꾸는 것도 쉽지 않다. 와이어트는 “처음부터 새로 구축한다면 전혀 다른 방식으로 설계할 수 있는 업무가 많지만, 레거시 워크플로우는 변화의 동력을 확보하기 어렵다”며 “재설계에는 상당한 시간과 비용이 필요하다”고 말했다.

그는 선도적인 CIO들이 이러한 문제를 극복하는 방식도 다른 경영진과 크게 다르지 않다고 설명했다. 변화의 필요성을 입증하는 비즈니스 케이스를 수립하고, 달성하고자 하는 목표를 명확히 하며, 그 결과가 가져올 가치를 구체적으로 제시해 필요한 자원을 확보한다는 것이다.

또한 “기회가 생길 때마다 워크플로우를 재정비하는 접근도 병행한다”고 덧붙였다.

로스 티스노프스키는 이러한 흐름이 이어지면서 CIO들이 점차 더 복잡한 영역으로 혁신을 확대하게 될 것으로 내다봤다. 그는 인프라 운영과 IT 지식 관리 체계 등 고도화된 워크플로우 영역에서도 변화가 본격화될 것이라고 전망했다.

직원 주도형 업무 혁신 확산

기술 기업 베이전의 CIO 패트릭 필립스는 AI만으로는 최대 효율을 달성할 수 없다는 판단 아래, 프로세스 개선 경험과 직원들의 현장 인사이트를 결합해 IT 워크플로우 혁신에 나서고 있다.

그는 “기존 프로세스에 AI를 덧붙이는 방식이 아니라, 프로세스 자체를 완전히 다시 정의해야 한다”라며 “만약 지금 AI 네이티브 도구로 처음부터 프로세스를 설계한다면 어떤 모습일지를 고민하고 있다”고 말했다.

필립스는 반복적이고 표준화된 업무가 많은 워크플로우를 중심으로, 직원들이 직접 혁신 대상 영역을 발굴하도록 했다.

그는 “직원들이 자신의 업무 방식을 어떻게 바꿀지 고민할 것으로 기대한다”며 “이를 위해 필요한 도구와 교육, 그리고 워크플로우를 처음부터 다시 설계할 수 있는 권한을 제공하는 것이 우리의 역할”이라고 설명했다.

대표 사례로 헬프데스크 조직이 있다. 필립스는 이 팀에 AI 기반 코드 에디터 ‘커서(Cursor)’를 도입하고, “이상적인 헬프데스크를 직접 설계해보라”고 주문했다.

필립스는 “직원들은 무엇이 불편한지 이미 알고 있었고, 업무를 더 쉽게 만들 동기가 충분했다”라며 “단순한 비밀번호 재설정보다 더 가치 있고 흥미로운 일을 하고 싶어 하기 때문에 스스로 효율성을 높이려 한다”고 말했다.

그 결과 헬프데스크 팀은 워크플로우를 재구성해 효율을 높였고, 확보된 시간을 기획 회의 참여 등 보다 부가가치가 높은 업무에 투입할 수 있게 됐다.

지속적 개선 문화로 전환

데이터 보호 및 사이버 복원력 플랫폼 기업 컴볼트의 CIO 하 호앙은 이러한 변화가 CIO에게 필수 과제라고 강조했다.

그는 “그동안 CIO들은 매출, 재무, 고객 지원 등 ROI가 명확한 비즈니스 워크플로우 혁신에 집중해왔다”라며 “하지만 IT는 상대적으로 소외된 영역이었다. 이제는 달라져야 한다”고 말했다.

이어 “IT가 조직의 혁신을 주도하려면 스스로 모범을 보여야 한다”며 “자동화와 AI, 효율성을 강조하면서 정작 내부는 티켓 처리, 수작업 중심 프로세스, 반복적인 업무 전환에 묶여 있어서는 안 된다”고 지적했다.

그는 “따라서 CIO들은 내부 IT 워크플로우에도 동일한 수준의 엄격함을 적용해야 한다”며 “이는 신뢰 확보와 비용 효율성을 동시에 달성할 수 있는 가장 빠른 방법”이라고 강조했다.

하 호앙과 그의 팀은 AI, 생성형 AI, 그리고 에이전트 기반 기능을 계기로 단순 최적화를 넘어 워크플로우 자체를 근본적으로 재검토하고 있다.

그는 “과거에는 ‘이 프로세스를 어떻게 더 빠르게 만들 것인가’를 고민했다면, 이제는 ‘이 프로세스가 왜 존재하는가’를 묻는다”고 설명했다.

이러한 접근은 다양한 영역에서 변화를 이끌고 있다.

우선 헬프데스크부터 혁신을 시작했다. AI 기반 셀프서비스와 가상 에이전트를 도입하고, 티켓 분류와 라우팅을 자동화했으며, 반복적인 문제는 자동으로 해결되도록 했다. 그 결과 IT 인력에게 전달되는 티켓 수가 줄고, 처리 속도는 빨라졌으며, 업무는 사후 대응 중심에서 보다 가치 중심으로 전환됐다.

이후 핵심 IT 워크플로우 전반으로 혁신을 확대하고 있다.

예를 들어 정책 기반 접근 권한을 자동으로 부여하는 체계를 구축해 수작업 승인 절차를 최소화하고 있다. 또한 데이터와 AI를 활용해 위험도가 낮은 변경 작업을 간소화하고, 변경 관리 과정의 병목을 줄이고 있다. 아울러 AI 기반 검색과 어시스턴트를 도입해 지식 관리와 문제 해결 과정에서 발생하는 사일로를 제거하고 대응 속도를 높이고 있다.

하 호앙은 이러한 변화가 시작에 불과하다고 강조했다.

그는 “이제 워크플로우 혁신은 일회성이 아니라 지속적으로 이어지는 활동이 됐다”라며 “가장 큰 변화는 기술이 아니라 사고방식의 전환”이라고 말했다.
dl-ciokorea@foundryco.com

• 관련 기사 : “정답 제시 대신 질문하고 판단하라”…AI 시대 필요한 리더의 조건을 묻다

IDC estimates there were over 28 million AI agents deployed by the end of last year, and predicts there’ll be over 1 billion actively deployed by 2029, executing 217 billion actions per day.

It’s easy to build an AI agent POC, says Venkat Achanta, chief technology, data, and analytics officer at TransUnion, a global credit reporting company with $4.6 billion in revenues. But governing, securing, and scaling it are a whole other challenge, especially for companies in highly regulated industries such as financial services and healthcare.

To address the problem, TransUnion spent the last three years building its agentic AI platform, OneTru. The goal was to make something as reliable and deterministic as the old, scripted, expert-style systems but as flexible as gen AI, and as easy to interact with as a chatbot.

The trick, however, was to combine the best of both worlds by using old-school systems for core processes where explainability and reliability are key, and layering in gen AI functionality in limited ways for the tasks it was uniquely suited for. And since the infrastructure to do this wasn’t available, TransUnion built its own, allocating $145 million to the project.

That was a big investment in an unproven technology, but it’s already led to $200 million in cost savings. More than that, once the platform was built, TransUnion used it to build customer-facing solutions.

In March this year, for example, TransUnion released its AI Analytics Orchestrator Agent, built using the OneTru platform and powered by Google’s Gemini models. The agent is already being used by TransUnion internally to improve analytics, and can also be used by customers to run sophisticated data analysis without the need for data scientists.

Many clients use TransUnion’s data but don’t use other solutions and platforms, Achanta says. The new orchestrator agent has the potential to help customers get more value out of the data, and unlock new revenue streams for the company.

And more agents are in the works, Achanta says. The key to making them work is the orchestration, governance, and security layers. Just making an agent do something is very easy for anyone, he says, and can take just a few days. The company can also create agents quickly. “But I have the foundation and guardrails, and the agent sitting on my platform uses all of them,” he says. “That’s what gives us power.”

The secret to making AI agents behave is to separate the layers of the task and assign each layer to a different system, each one operating under a set of constraints. This approach limits the damage any particular agent can do, creates a system of checks and balances, and restricts the riskiest activities to a pre-gen AI technology.

For example, at TransUnion, the core decision-making is performed by an updated version of an expert system. It operates under a set of well-defined, auditable rules and works predictably, cost-effectively, and at low latency. When it encounters a situation it hasn’t seen before, an LLM is used to analyze the problem, a different agent might then turn it into a new rule, and then a human might be called in to review the results before the new rule is added to the expert system. There are different agents that understand the semantic layer, interact with humans, and perform other tasks.

“With the neural reasoning layer — the LLM — we put humans in the loop,” he says. “When it’s a symbolic reasoning layer, which is logic and machine-learning-driven, we let it be automated.”

So when each agent operates within very narrow constraints, on just the limited data it needs for that one task, and is limited to what it can do, the entire system becomes much more governable and reliable.

It’s like the difference between an assembly line, where multiple workers each do a single, distinct task, instead of a workshop where a single artisan does everything. The assembly line can do work faster and more reliably but today, many enterprises deploy their AI agents as if they were craftsmen. The latter approach can result in creative, unique products, but this isn’t always what a company needs.

Nicholas Mattei, chair of the ACM special interest group on AI and professor at Tulane University, suggests that companies focus on building in extra security at points where different parts of the agentic system connect.

“Make sure you have security at the seams,” he says. For example, if an agent sends requests to an email service, set up a checkpoint between the two. “Around the gaps between the unreliable agents and where the traditional software lives, that’s where you want to focus your security processes,” he says.

Building a security foundation for agentic AI

In a Jitterbit survey of 1,500 IT leaders released in March, AI accountability — security, auditability, traceability, and guardrails — is the biggest factor when it comes to the final AI purchase decision, ahead of speed of implementation, vendor reputation, and even TCO. Security, governance, and data privacy risks were also top issues preventing AI initiatives from moving to production, ahead of costs and integration challenges. And they’re right to be worried.

Earlier this year, researchers at cybersecurity firm CodeWall were able to breach McKinsey’s new AI platform, Lilli. Using an AI tool of their own, the researchers said they could access 47 million chat messages, 728,000 files, 384,000 AI assistants, 94,000 workspaces, 217,000 agent messages, nearly 4 million RAG document chunks, and 95 system prompts and AI model configurations.

“This is decades of proprietary McKinsey research, frameworks, and methodologies — the firm’s intellectual crown jewels sitting in a database anyone could read,” the researchers wrote.

The reason? Out of over 200 publicly exposed API endpoints, 22 required no authentication. It took just two hours for the researchers to get full read and write access to Lilli’s entire production database. McKinsey responded quickly to the alert, patched the unauthenticated endpoints, and took other security measures.

“Our investigation, supported by a leading third-party forensics firm, identified no evidence that client data or client confidential information were accessed by this researcher or any other unauthorized third party,” the firm said in a statement.

IDC says the incident underscores just how dangerous the breach of an AI system can be to an enterprise.

“Most companies are still thinking about AI risk in yesterday’s terms: data leakage, bad outputs, and brand reputation damage,” says Alessandro Perilli, IDC’s VP for AI research. “Those are serious issues, but the bigger risk becomes delegating authority to AI systems.”

By getting access to an agentic AI platform, an attacker can’t just see something they’re not supposed to, but also covertly change how the company acts. And securing enterprise-scale agentic AI systems like Lilli is only half the challenge. According to Gartner, 69% of organizations suspect employees use prohibited AI tools, and 40% will experience security or compliance incidents by 2030 as a result.

But available discovery tools aren’t fully ready to find AI agents, Gartner says.

“If I asked you how many agents run in your enterprise right now, where are you going to go look it up?” asks Swaminathan Chandrasekaran, global head of AI and data labs at KPMG, which now has several thousand AI agents in production. “Have they all been onboarded and have identities? Have they gone through a proper authentication process and who’s in charge of them? That piece of infrastructure doesn’t exist.”

Tools are just starting to emerge, however, or companies are creating DIY solutions, he says. “That’s what’s going to give CIOs peace of mind,” he says.

We’re already seeing public examples of individual employees deploying powerful agentic AI to negative consequences. Summer Yue, Meta’s alignment director, recently decided to use OpenClaw, a viral open-source agentic AI tool, to help handle her inbox. After it worked in a test inbox, she deployed it for real.

“Nothing humbles you like telling your OpenClaw to confirm before acting and watching it speedrun deleting your inbox,” she wrote on X. “I couldn’t stop it from my phone. I had to run to my Mac mini like I was defusing a bomb.”

In the past, an employee might upload sensitive information to a chatbot or ask it to write a report that they’d then copy and paste, and pass off as their own. As these chatbots evolve into full-on agentic systems, the agents now have the ability to do anything a user has privileges to do, including accessing corporate systems.

To manage this new security risk, companies will need to move past role- and identity-based controls to intent-based ones, says Rakesh Malhotra, principal in digital and emerging technologies at EY.

It’s not enough to ask whether an agent has permission to access a system to make a change to a record, he says. Companies have to be able to ask why are you changing this. That’s a big challenge right now.

“The observability stacks don’t capture the intent of why the agent did something,” he says. “And that’s really important to understand. Trust is based on intent, and there’s no way for any of these systems to capture intent.”

If a human employee tries refactor the entire code base, they’d be asked to provide a good reason for doing that. “And if you’re refactoring without any specific reason, maybe you shouldn’t do it,” Malhotra says. “With people, there are ways for this to be adjudicated. I don’t know how to do this with agents.”

Building a semantic data foundation for agentic AI

TransUnion’s Achanta repeatedly mentioned the semantic foundation of the company’s OneTru platform. Such an understanding of information helps systems understand not just what the data is, but what it means, and how it relates to other data. Gartner says developing a semantic layer is now a must-do for companies deploying AI.

“It’s the only way to improve accuracy, manage costs, substantially cut AI debt, align multi-agent systems, and stop costly inconsistencies before they spread,” the firm says.

By 2030, universal semantic layers will be treated as critical infrastructure, alongside data platforms and cybersecurity, Gartner predicts. And agents need context to be able to do anything meaningful with data, says KPMG’s Chandrasekaran. That’s where a company’s knowledge is contained.

“That’s your new IP for the enterprise,” he says. “Context is the new moat.”

For John Arsneault, CIO at Goulston & Storrs, creating a solid data foundation is also a way to avoid vendor lock-in.

“If you’re buying things and moving your data into them to create workflow automation or agentic work assistants, you’ll have a hard time getting out of it,” he says. “But if you take a data-centric approach, you can at least move from one to the other if there’s a shift in the marketplace.”

The law firm has migrated its client-oriented work products into NetDocuments, a document management system specifically focused on the legal industry. And for the rest of the data the company collects, it goes into Entegrata’s legal data lakehouse.

“Our goal is to have all our other applications eventually point at that data lake,” he says. “Then we’ll have these two environments where all the firm’s data exists, which will allow us to put any AI tool we use on top.”

It’ll also make the data flows easier to manage, he adds, and will enable the firm to adapt quickly to whatever AI technology comes next. “Whether gen AI, agentic, or Anthropic stuff, with the Cowork legal plugin, it’s very difficult to keep up with,” he says. “And it changes every six months.”

Agentic orchestration

The last part of the agentic infrastructure puzzle, after getting security guardrails in place and creating a usable data layer, is orchestration. Agentic AI systems require agents talk to each other and human users, and interact with data sources and tools. It’s a complicated challenge, and this technology is still very much in its infancy, though moving quickly. MCP is one such example, and is a key piece of solving the orchestration puzzle. AI vendors have been remarkably willing to cooperate here.

“When social networks were born, and Facebook and Twitter were discussing a standard protocol for interacting, nobody wanted to adopt their competitors’ protocol,” says Agustin Huerta, SVP of digital innovation and VP of technology at Globant, a digital transformation company. “Now everyone is going through MCP and maturing it as a standard protocol.”

But that’s not to say agentic integration has been solved. According to a Docker survey of more than 800 IT decision makers and developers, the operational complexity of orchestrating multiple components is the biggest challenge when it comes to building agents.

In particular, 37% of respondents say orchestration frameworks are too brittle or immature for production use, and 30% report testing and visibility gaps in complex orchestrations.

In addition, while 85% of teams are familiar with MCP, most say there are significant security, configuration, and manageability issues that prevent deployment in production. And there are other integration issues enterprises have to deal with.

“One problem yet to be solved is how to get a proper dashboard to control all these agents, to know exactly what’s going on with each of them,” says Huerta. “One dashboard will let you monitor agents built with OpenAI, and one is for agents that live on Salesforce, but none can expose telemetry in a central dashboard for control, auditing, and logging.”

For companies just starting to deploy agents, or who are sticking to a single platform, this isn’t yet an issue, he adds, but as they leverage a larger network of agents, they’ll start to experience the challenges. Globant itself is building its own internal dashboard for agentic AI, for instance.

And at Brownstein Hyatt Farber Schreck, a 50-year-old law firm with about 700 employees and clients around the US, there are several areas where AI is being deployed, including a proposal generator system.

Normally, it can take several people days to review a client’s request for proposal, go through hand-written notes or meeting transcripts, and pull together other relevant materials, says Andrew Johnson, the firm’s CIO.

“We can feed all that information into a computer and extract key criteria to produce a quality first draft in minutes,” he says.

Multiple agents are required for different parts of the process — one to extract success criteria or staffing requirements, one to look for precedents and lessons learned, and others for pricing and the brand standards. “Each of those agents is autonomous and needs to be orchestrated so the outputs of each are fed into the next step,” Johnson says. For the most part, that means a RAG system, since most of the legacy platforms the firm uses have yet to incorporate an MCP layer.

Depending on the task, individual agents may be powered by different models, which is another layer of orchestration that needs to be managed.

Then there’s cost monitoring. If an AI agent or group of agents gets into an infinite feedback loop, the inference costs can quickly rise.

“We’re aware of the concern, though we have yet to see it manifest,” says Johnson. “So we have monitoring in place. If we exceed thresholds, we react to it.”

Regardless of strategies or measures to absorb setbacks, everything having to do with AI is changing faster than anything else companies have seen.

“I’ve been in technology for 25 years and I’ve never seen anything like this,” says EY’s Malhotra. “The fastest growing companies in the history of companies have all been created in the last three to four years. The growth in adoption is just unprecedented. And I talk to clients all the time implementing technologies that were highly relevant nine or 10 months ago, and everyone’s moved on.”

Mike Anderson gave his IT workers an unusual task: Create a Gemini Gems digital twin of your role, feeding the AI details such as technical documentation to learn about the role’s tasks and knowledge needs.

Anderson wants these AI creations to assist the human workers in doing their jobs by helping them access the right information in near real-time with just a query.

“We created these Gems that act as experts our team can lean on, and my goal is for each worker to get some time back,” says Anderson, chief digital and information officer at security tech company Netskope.

It’s just one of many ways Anderson is trying to transform workflows and processes inside his department to boost its efficiency and productivity. And he’s seeing wins.

For example, development teams are using AI to generate code. Workers use vibe coding to quickly develop “something to iterate on,” slicing months of work off a typical product development schedule, Anderson says. As part of the transformation, Anderson’s team is creating primitives to ensure certain elements, particularly security controls, will always be part of the AI-generated code, another workflow transformation to save IT workers time.

Although Anderson hasn’t calculated a detailed ROI for such efficiency gains, he says it’s clear he can do more with less as a result. “I’m able to maintain a flat budget and deliver more things and more capabilities than I could before,” he says.

‘Do more with the team we have’

CIOs have long felt the pressure to do more with less. And they’re feeling that pressure acutely today. According to survey findings from research firm Gartner, 57% of CIOs face pressure to improve productivity and 52% to reduce costs.

CIOs also have a mandate to use technology, especially AI, to transform workflows across the business to deliver productivity and efficiency gains. They also are seeking to deliver as much in their own IT departments.

“We want to make sure we’re daring to reimagine IT processes and what’s possible with AI to unlock value,” says Anisha Vaswani, chief information and customer officer at Extreme Networks.

Vaswani has put transformation of IT workflows front and center. Like Anderson, she’s using AI (specifically Claude Code) to speed coding. She has adjusted workflows, shifting workers from writing code to prompting, reviewing, and managing quality.

She is also transforming help desk operations, as many CIOs are, using AI and automation to increase self-service options.

More complex workflows are benefitting as well, says Vaswani, who is using AI to scale IT’s QA function by generating test strategies and automating testing. “There is a lot of promise there to reduce to minutes what could take weeks manually,” she says. She sees AI’s use as a way to add capacity without increasing costs.

And she’s exploring how IT teams can use AI to better capture user requirements when working on new products or enhancements.

“I want to reimagine how IT interacts with our business partners to be more responsive and agile, so we can deliver value more frequently and be a lot more customer-centric,” she adds. “The goal is to do more with the team we have. We want to innovate faster, to deliver more, to get more done.”

Transformational mandate

Alex Wyatt, a director at consultancy West Monroe, says IT’s process-driven work is ripe for transformation.

“AI has re-sparked the conversation around that,” he says. “Now with AI CIOs are getting more pressure to cut costs, and boards are saying, ‘This process has to get 50% more efficient.’”

As is the case throughout a typical organization, CIOs can — and should — go after the easier transformation tasks first to chalk up wins and build skills before tackling harder goals, Wyatt advises.

“There are different stages with optimizing your workflows and processes,” he says.

The first stage is using AI to automate repetitive tasks and shift humans to oversight functions. The stage also involves maximizing the capabilities that system providers have incorporated into the tools and technologies that IT uses to get work done.

“That’s the lowest hanging fruit and where you can get the biggest bang for your buck in the shortest amount of time,” Wyatt says. “And then you go after the more sophisticated opportunities that exist.”

Although Wyatt says AI has pushed workflow and process transformation into the limelight, “it doesn’t mean AI is the only solution.” He reminds executives that “there are opportunities for pure process improvement,” like lean process design.

“There is a risk of automating a bad, inefficient process, so you have to think about outcomes, the KPIs you’re driving, and how you restructure work to achieve those,” he adds. “You don’t get more efficient by buying more tools. You have to rethink your workflows. You have to think about how people work.”

That longstanding advice has proven its worth through the decades. Aggressive workflow and process redesign can deliver significant gains — well over 50% improvements — “but to get that you really have to think about how work gets done and have the systems in place to do that,” Wyatt says.

Not all CIOs are able to put in the amount of work it takes. That shouldn’t deter CIOs from doing something, he adds. “You can get a 10% to 20% lift from doing some basic process redesign and then use AI to get you even more lift,” he notes.

Ross Tisnovsky, a partner at Everest Group and leader of the firm’s CIO research and advisory practice, stresses the transformation component, saying CIOs can encounter problems if they automate without it.

Efficiency gains in coding/development and testing are case in point, he says, noting that AI is boosting efficiency in coding by 70% or more, while on the testing side it’s closer to 30%. As a result, CIOs who don’t remake workflows soon find they’ve created an imbalance, with code being produced faster than testing can handle.

As such, Tisnovsky suggests, the reason most AI initiatives don’t deliver value often stems from failure to rework the workflows.

Dual pressure and desired outcomes

CIOs face other challenges in transforming how the IT department works, Wyatt says.

To start, they and their IT teams are fully engaged in doing such work for all the other departments in the organization, with priority typically given to transformations that boost revenue, market share, customer retention, and the like.

“IT is in a unique position: They’re being asked to do transformation across the organization as well as in IT. They have that dual pressure,” he says.

That can challenge CIOs’ ability to allocate the resources and skills needed to redesign IT workflows, Wyatt says.

CIOs also find it hard to get rid of embedded workflows. “There is a lot of work that if you were to build from scratch, you’d do it a different way, but with legacy workflows it is hard to get the momentum to change. It takes a lot of time and money to reengineer,” Wyatt explains.

He has found that leading CIOs overcome such challenges the same way other execs do: putting together business cases for the transformation, focusing on desired outcomes, articulating the value those outcomes will deliver, and using all that to get resources to do the work.

“And they look to rework workflows as opportunities present themselves,” Wyatt adds.

Tisnovsky says as all that happens, CIOs will begin transforming more complex workflows, such as those in the infrastructure space and in the IT knowledge base.

Empowering workers to transform their tasks

Patrick Phillips, CIO of tech company Vasion, is drawing on his process-improvement experience and his workers’ insights to transform his IT department’s workflow, knowing AI alone won’t create maximum efficiencies.

“The process itself has to be completely redefined, so you’re not bolting AI onto legacy processes,” he says. “So I’m asking what if we were building the process today with AI-native tools, what would it look like?”

Phillips has tasked his workers to identify workflows that are primed for transformation, such as those filled with commodity-level tasks.

“Our expectation is that they’re going to consider transforming how they work, and we have the obligation to help them do that by providing the tools and trainings and the empowerment for them to rebuild workflows from ground up,” he says.

For example, Phillips enlisted help-desk staff to transform how that team operates, equipping them with Cursor, an AI-powered code editor designed for software development, and challenging them “to build the help desk just like you think it should be built.”

“They knew what bugged them, and so they had the incentive to make their jobs easier,” Phillips says. “They want to make themselves more efficient because they want to do things that are more valuable and more interesting than resetting a password.”

Phillips says the help-desk staff reworked workflows, boosting efficiency and enabling them to shift time to other responsibilities such as participating in planning meetings.

Instituting a culture of continuous improvement

Ha Hoang, CIO of Commvault, maker of a data protection and cyber resilience platform, says taking on such work is an imperative for CIOs.

“CIOs have historically been very focused on transforming business workflows — sales, finance, customer support — because that’s where the visible ROI is,” Ha says. “But IT has often been the cobbler’s children. That’s changing now, and it needs to. If IT is going to lead transformation, it has to model it. You can’t be pitching automation, AI, and efficiency to the business while your own teams are buried in tickets, swivel-chair processes, and manual handoffs.

“So yes,” she adds, “CIOs should absolutely be putting the same rigor on internal IT workflows. In many cases, that’s actually the fastest path to credibility and cost efficiency.”

AI, generative AI, and now agentic capabilities have prompted Ha and her IT team to rethink workflows entirely, not just optimize them.

“Before AI, we were asking, ‘How do we make this process faster?’ Now we’re asking, ‘Why does this process even exist?’” she explains.

That has led to transformations in various areas.

As is common, Ha and her team started with transforming help-desk workflows, where they have deployed AI self-service and virtual agents, automated ticket triage and routing, and enabled auto-resolution for common issues. The results are fewer tickets reaching IT staff, faster resolution times, and a shift from reactive support to higher-value work.

Then they started looking for transformation opportunities across core IT workflows.

For example, they’re moving to fully automated, policy-based access provisioning with minimal manual approvals. They’re using data and AI to streamline low-risk changes and reduce bottlenecks in change management. And they’re deploying AI-powered search and assistants to eliminate silos and speed issue resolution in knowledge work and troubleshooting.

Ha says these changes are just the start.

“For me, it’s become a continuous discipline, not a one-time initiative,” she says, noting that might be one of the most significant work transformations. “I think the biggest shift isn’t technology, it’s the mindset.”

In conversations I’ve had with CIOs over the past year, there’s been a noticeable shift in how NIS2 (Network and Information Security Directive 2) is being discussed. It used to be filed away as another regulatory hurdle to clear, but now it’s prompting CIOs and their teams to think a little deeper about how well they understand the systems they depend on. For a long time, risk has been largely framed within the boundaries of the organization — something that could be managed through internal controls, policies and audits. But that no longer reflects how digital services are built or delivered. Most organizations I encounter rely on a web of providers spanning cloud platforms, data centers, network operators and software vendors, all working together to create a “patchwork” ecosystem. NIS2 is different because it acknowledges that reality and, in doing so, it’s forcing a broader and sometimes more uncomfortable reassessment of where risk really sits.

What stands out to me is that NIS2 doesn’t just focus on individual accountability, but on the very definition of resilience itself. It recognizes that disruption rarely originates within a single process, or even a single organization. More often, it emerges from the connections between them; from unseen dependencies, indirect relationships and assumptions about how systems will behave under pressure. That’s novel, because it moves the conversation away from whether individual systems are secure, and toward whether the overall architecture those systems sit within can continue to function when something inevitably goes wrong. In that sense, NIS2 is less about tightening cybersecurity controls and more about encouraging a different way of thinking, where resilience is shaped as much by how infrastructure is designed and connected as it is by how it is protected.

NIS2 expands the definition of risk beyond the enterprise

One of the most immediate impacts I’m seeing from NIS2 is how it challenges long-held assumptions about control. Speak to any CIO, and they’ll usually talk about securing what sits within their own environments — their applications, services and data. But in practice, very little of today’s digital estate is fully owned because it’s so distributed among third parties with countless links and dependencies. Virtually all business services depend on layers of external providers, each with its own dependencies, architectures and risk profiles. According to the World Economic Forum, the top supply chain risk in 2026 is the inheritance risk — the inability to ensure the integrity of third-party software, hardware or services. NIS2 brings that into sharp focus by extending accountability beyond direct suppliers to include the wider ecosystem that supports them. In essence, it prompts businesses to shift from asking “are we secure?” to “how secure is everything we rely on to operate?”

That’s quite a challenge, because it’s not enough for businesses to simply know their suppliers — they need to understand how deeply interconnected those relationships are. In many cases, the real exposure sits several steps removed, in the providers behind your providers or in shared infrastructure that underpins multiple services at once. The “uncomfortable reassessment” I mentioned earlier is the squaring of this circle — how many organizations have full visibility into that sprawling landscape, let alone the means to control it?

NIS2 is compelling organizations to map dependencies more rigorously, to ask harder questions of their partners and network infrastructure, and to recognize that resilience is only as strong as the most fragile link in the chain. The WEF shows that in 2026, only 33% of organizations map their entire IT supply chain to gain this visibility. And even then, the added risk of unknown service providers, such as is the case when suing the public Internet, where data pathways are neither visible nor controllable, is difficult to quantify.

Compliance is the trigger, but architecture is the challenge

What I find interesting about NIS2 is that it goes deeper than compliance — it’s trying to trigger a shift in culture. It’s relatively straightforward to introduce new policies, expand reporting requirements or formalize supplier assessments. But what happens when those requirements collide with the reality of how modern IT environments are built? Many organizations simply don’t have a clear, end-to-end view of how their services are delivered, how data flows between providers or how incidents might spread like wildfire across the ecosystem they depend on. NIS2 asks CIOs to look beyond governance frameworks and examine whether their operating models support the level of oversight and responsiveness the directive expects.

And that is where the architecture question becomes essential. It’s one thing to require suppliers to report incidents or meet certain security standards; it’s another thing entirely to ensure that the underlying infrastructure is designed to absorb disruption without cascading failure. In my experience, this is where many organizations begin to realize that resilience cannot be layered on afterwards. It must be built into how systems are structured, how dependencies are managed and how connectivity is established between environments. NIS2 may define what needs to be done, but it doesn’t prescribe how to do it. That responsibility sits with CIOs, who now have to translate regulatory intent into practical design decisions about where workloads run, how services interconnect and how failure is contained when it occurs.

Infrastructure design is now resilience design

What this ultimately leads to is a big infrastructure rethink. I’m privileged to have had some interesting discussions with CIOs and other executives about this very topic, so I know that resilience is beginning to be understood as more than a set of security controls. Connectivity is now at the heart of resilience, and in that sense, NIS2 has succeeded in getting organizations to think differently about what resilience really means. If a service depends on a single cloud region, a single network path or a tightly coupled set of providers, then no amount of policy or monitoring will prevent disruption when one of those elements fails. I’m pleased to see organizations starting to question these assumptions — not just asking whether systems are secure, but whether they are structured in a way that allows them to continue operating under stress. That shift in thinking does away with the abstract theory of resilience and defines it as something that can be designed and architected.

From a connectivity perspective, this means building in diversity at every level. Distributing workloads across geographically separate locations, establishing multiple, independent network paths and avoiding unnecessary concentration of critical services all contribute to a more resilient architecture. Interconnection plays a starring role here as the mechanism that allows different parts of the digital ecosystem to communicate in controlled, redundant and predictable ways. When designed properly, this kind of architecture limits the blast radius of any single point of failure and makes it easier to maintain service continuity even when parts of the system are down or under strain. The real takeaway here is that resilience is not something any single organization can achieve in isolation. It emerges from the collective design of the entire ecosystem, where each participant contributes to the overall stability of the services they all depend on.

When regulatory pressure gives way to strategic opportunity

The building blocks are already there. Practices like supplier due diligence, security certifications and business continuity planning are not new. What NIS2 does is raise the bar on how consistently and how deeply they are applied. It also brings a level of structure to conversations that were previously fragmented, particularly when it comes to expectations between partners. And therein lies the strategic upside. Organizations that can clearly demonstrate how they manage risk across their supply chains, how they design for resilience and how they respond to disruption are in a stronger position, not just from a regulatory standpoint, but in how they engage with customers and partners. In some sectors, we’re already seeing this play out through increased requests for transparency, self-assessments and proof of compliance. That trend is only going to accelerate. For CIOs, it’s a golden opportunity to move beyond a defensive posture and position resilience as a key competitive differentiator. It becomes a way to build trust, strengthen relationships and support more sustainable growth, rather than simply a requirement to satisfy regulators.

NIS2 may be the catalyst, but the underlying change runs deeper. It’s pushing CIOs to think beyond compliance and toward a more structural understanding of risk that reflects how digital services operate today.

This article is published as part of the Foundry Expert Contributor Network.
Want to join?

There’s been a wake-up call for CIOs. All the talk about perceived productivity boosts that have previously dominated conversations about AI has been replaced with a demand for measurable value from investments in emerging tech.

As MIT states that project failure rates are as high as 95%, executive boards are starting to question when AI will pay dividends. PWC’s Global CEO Survey shows that more than half of companies have seen neither higher revenues nor lower costs from AI, and only one in eight have achieved positive outcomes.

While Gartner predicts significant growth in AI spending this year, John-David Lovelock, distinguished VP analyst at the research firm, says the lack of tangible returns means digital leaders are changing tack. Rather than hoping their AI explorations will produce returns, CIOs are switching to more targeted initiatives.

“The projects growing quickly are the ones doing business, and those initiatives include AI,” he says. “CIOs are starting to de-emphasize AI and re-emphasize business. These projects are about AI enhancing existing work and moving away from moonshot transformational projects.”

Lenovo’s CIO Playbook for 2026, produced with tech analyst IDC, also suggests enterprises will get serious about AI deployments this year, with explorations replaced by production-level services that drive business transformation. With boards exerting pressure for measurable returns, Ewa Zborowska, research director at IDC, says more digital leaders want to use AI to enhance, innovate, and reinvent their organizations.

“CIOs aren’t just considering AI out of curiosity, they want to see what they can get out of it to grow the business,” she says. “AI adoption is much more about doing new things or taking a fresh approach to creating value rather than becoming more efficient at cost-cutting.”

Such is the clamor for value that Richard Corbridge, CIO at property specialist Segro, suggests that returns from AI are a main digital leadership priority: “If you discover, for example, that everyone in the organization used Copilot 10 times today, that might mean they’ve been more efficient,” he says. “But what have they actually done with the time they saved? How has saving time created value?”

CIOs will grapple with these questions during the next 12 months. With CEOs and boards becoming impatient for returns, digital leaders are working more with their bosses to define value. Successful CIOs fine-tune their arguments to ensure their projects are backed, and then demonstrate the value of their AI initiatives to the board.

Defining a valuable AI project

What’s clear is CIOs can’t deliver outputs from AI projects without input from their enterprise peers. IDC’s Zborowska says tighter cooperation across project ownership and KPIs ensure emerging technology investments are targeted at the right places.

This increased interaction between digital and business leaders also changes project aims. As stakeholders work closely together to generate value from AI, Zborowska expects executives to seek KPIs that stretch across operational concerns.

“I’d bet we see more non-financial aims over the next few years,” she says. “Executives will consider things such as are employees more engaged, has their work improved in any way, are AI implementations impacting customer experiences, and are internal decisions being made more efficiently.”

Martin Hardy, cyber portfolio and architecture director at the UK’s Royal Mail, agrees that defining valuable AI projects is all about finding the right focus. Effective deployments target processes in distinct areas, and business stakeholders must be part of the value-defining process.

“If we’re making decisions about legal documentation, AI is probably not there yet,” he says. “But if we can use AI to approve holidays, for instance, that might be something because if you have rules that say no more than two people off at a time, you could use AI to check about booking holidays without having to ask everyone in the office.”

For CIOs seeking value-generating use cases, Gartner’s Lovelock suggests AI can deliver results in key business areas such as boosting revenue, supporting decision-making, engaging staff, and improving experiences. He says the right path to AI exploitation correlates with Gartner’s enterprise technology adoption profiles, which group companies into a range of categories.

“The folks who are furthest forward, what we call the agile leaders in technology, are much more likely to drive AI to change the business,” he says. “The laggards on the other side are more likely to take on the technology that’s given to them by incumbent software providers, and use it in a prescriptive manner.”

Fine-tuning the use case

The challenge now is for digital leaders to work with their business peers to determine a more refined approach to AI deployment. For some CIOs, the value of AI is clear but the potential risks must be considered.

Take Dan Keyworth, executive director of performance technology and systems at McLaren Racing, whose focus is operational stability and race-day reliability. While he says being aware of developments in generative and agentic AI is crucial, the priority is tried-and-tested technologies rather than innovations that put performance at risk.

“Formula One is grounded in traditional machine learning and simulation,” he says. “Developing models has been a big part of our performance journey, and since the engine already existed, gen AI is the turbo that’s bolted on with more investment in AI.”

For other digital leaders, like Barry Panayi, group chief data officer at insurance firm Howden, success depends on keeping the human in the loop. Yes, automation can improve customer service, but rather than replacing staff, he wants to use AI to ensure Howden’s professionals have the right insight when they interact with clients.

“There’s absolutely no desire to use data to drive productivity by automating what we do with our customers,” he says. “This is a business where people speak to people. Our brokers need information that can give them an edge, and prove to their clients they understand the risks and can give them the best deals.”

Nick Pearson, CIO at technology specialist Ricoh Europe, adds that the use case for AI at his firm is two-fold: boosting operational productivity and improving customer processes. So he’s established a tri-party AI council with the head of service operations and the commercial manager in Spain. This council explores opportunities to buy, build, and reuse emerging tech.

“We’ve got a strategy that looks at where AI matters, which means exploring the technology we already have to boost internal productivity,” he says. “We’ve got a lot of people who know how to code and build things in Copilot Studio and other platforms, so let’s use that to increase productivity.”

Showing returns to the board

For Gartner’s Lovelock, the key lesson for CIOs eager to generate value from AI is to work with their peers and set desired outcomes before investing. “Most people start with the idea that more is more, and if you do that, you won’t get to the idea of quality,” he says.

That sentiment resonates with Segro’s Corbridge, who encourages digital leaders to start conversations with other professionals by focusing on value. Ask people how investing in an AI implementation will create value for them personally, for the wider business, and the customers the organization serves.

He says CIOs shouldn’t try to prove that AI works, but rather concentrate on how emerging tech adds value. That definition is so critical to Segro’s way of working that the organization uses the phrase proof of value rather than proof of concept.

“Most things work, but they might be more expensive,” he says. “For example, you might be able to use AI to transform how the organization uses spreadsheets, but that project might cost you $300,000. And if you’re currently paying someone $40,000 to do that work, and they’re happy doing it, then you have to question the value.”

Lessons are being learned, says IDC’s Zborowska, whose firm’s research suggests that half of AI POCs now transition into production. While some people might think this success rate isn’t impressive, the quantity a year ago was 10%. After several years of AI exploration, it appears CIOs and their businesses are now firmly focused on real returns.

“These numbers speak to the fact that companies are being more mature and mindful in how they allocate budgets,” she says. “They also support the main theme that we’re on a journey to transformation and a maturing market for AI adoption.”

A widely shared Template22 graphic on why projects fail prompted this article. I am using that chart as a prompt, not as evidence. The more useful question is not whether the familiar causes of failure are real. They are. The more useful question is why they keep repeating across programs, portfolios and enterprise transformations, even after years of investment in methods, PMOs, digital tools and AI.

The answer, in many cases, is not a lack of effort. It is a lack of decision logic. Enterprises still launch, govern and defend large initiatives without a planning discipline capable of calculating trade-offs, exposing constraints, modeling dependencies and recalculating the impact of change quickly enough to support real governance.

The pattern under the pattern

Most discussions of project failure start with visible symptoms, unclear scope, weak requirements, scope creep, poor communication, resource shortages, unrealistic deadlines, weak sponsorship and poor change control. Those symptoms matter, but when they recur at scale, they usually point to a deeper problem in the planning system itself. In PMI’s 2025 research on the strategy execution gap, PMI President and CEO Pierre Le Manh argued that AI will create value only when organizations can translate bold ideas into executed initiatives. In most enterprises, the gap is not ambition. The gap is conversion. Strategy is declared, portfolios are funded, work begins, yet leaders still cannot calculate trade-offs, expose constraints, model dependencies or replan fast enough when conditions change.

The scale of the issue is hard to dismiss. BCG’s 2024 study of large-scale technology programs found that more than two-thirds are not expected to be delivered on time, within budget and within scope, and that only 30% fully meet expectations on those three dimensions. Gartner’s 2024 survey found that only 48% of digital initiatives across the enterprise meet or exceed their business outcome targets. Those are not isolated execution misses. They are signs of systemic underperformance in how organizations prioritize, fund, sequence and govern change.

Other firms sharpen the diagnosis from different directions. McKinsey’s work on successful transformations found that among companies whose transformations failed to engage line managers and frontline employees, only 3% reported success. Bain’s David Michels argues that “red is good,” meaning organizations perform better when risk is surfaced early rather than hidden behind reassuring dashboards. Deloitte’s research on digital acceleration and strategy makes the strategic requirement explicit: Digital possibilities must shape strategy, and strategy must shape digital priorities. Put together, those findings point to one conclusion. Large programs rarely fail because a single team misses a task. They fail because the enterprise cannot see the interaction of priorities, constraints, dependencies and consequences early enough to respond intelligently.

Why this is a planning problem, not just a delivery problem

At the portfolio level, failure begins when organizations select too much work, fund the wrong work or fund the right work without a realistic view of capacity, technical debt and delivery interdependencies. BCG ties poor outcomes directly to inaccurate timeline and resource planning, weak end-to-end roadmaps and ineffective management of interdependencies. That is not simply a delivery problem. It is a portfolio design problem. Forrester’s 2025 work on operating model change adds a related warning: Fewer than half of IT leaders say their organizations prioritize operating model adaptation, leaving strategy to collide with structures that are not built to absorb change.

At the governance level, failure shows up as a value problem. Traditional oversight mechanisms can collect status, enforce templates and schedule reviews, yet still fail to answer the executive question that matters most: What happens if a key dependency slips, a budget is reduced or a shared team becomes overcommitted? Bain’s “red is good” matters here because watermelon reporting, green on the outside and red underneath, is usually a sign that governance is reporting milestones instead of modeling consequences. Gartner’s survey of Digital Vanguard organizations reinforces the point. The highest performing digital organizations do better when business and technology leaders are more aligned on execution and outcome ownership.

At the execution level, the familiar problems remain, but they look different when viewed through a planning lens. PMI’s communications research found that one out of five projects is unsuccessful due to ineffective communication, and PMI’s later analysis of communication failures linked poor communication to more than half of the projects that fail to meet business goals. The important nuance is that communication is not merely a soft skill problem. It is often a failure to express the implications of planning decisions in a form that the business can act on. An unclear scope can be a weak scenario definition. Poor requirements can reflect commitments made before constraints were visible. Scope creep is often an unmanaged consequence. Weak sponsorship often reflects weak evidence. Poor change control often means the organization can log a change but cannot calculate its ripple effects.

Why algorithmic planning is now a governance requirement

This is where the conversation needs to become more precise. Continuous scenario planning is valuable, but it only becomes decision-grade when it is supported by algorithmic planning. In large programs and portfolios, governance cannot rely on static reporting, intuition or periodic review alone. It must be able to calculate the impact of change quickly, expose hard constraints clearly and place dependencies, capacity limits, sequencing conflicts and trade-off consequences where they belong, at the center of decision-making. Without that discipline, governance is mostly a matter of interpretation. With it, governance becomes evidence-based control. That conclusion follows directly from the documented failure patterns of PMI, BCG, McKinsey, Bain, Deloitte and Gartner.

AI makes this requirement even more important. Used well, AI can be a powerful interface for senior leaders, helping them interrogate scenarios, surface anomalies, summarize risks and engage more directly with the planning environment. Used badly, it can do the opposite. If AI is not tightly coupled to mathematically sound planning data, explicit constraints, dependency logic and algorithmic calculations, it can turn supposition into false confidence. That is dangerous in portfolio and program governance, where plausible-sounding answers are not the same as decision-grade answers. The sequence matters. First, the organization needs a locked down, calculation based planning model with clear borders. Then AI can sit on top of that model as an accelerator, interpreter and executive interface. Without those boundaries, AI can easily magnify weak assumptions rather than expose them. This caution is consistent with PMI’s strategy execution framing and with EY’s 2026 CEO Outlook and Accenture’s AI reinvention thesis, both of which insist that AI must be scaled with discipline and strong foundations.

Strategic intent is inherently directional. Governance must be exacting. The bridge between the two is algorithmic planning. It is the mechanism that translates ambition into modeled consequences by testing scenarios, exposing constraints, mapping dependencies and recalculating trade-offs as conditions change. Without that bridge, governance becomes subjective. With it, leadership can distinguish between what is desirable, what is feasible and what is now at risk. That is why constraints, dependencies and capacity should not be treated as soft considerations. They are the black-and-white rules of execution.

AI is most valuable when it explains a sound planning model, not when it improvises one.

Why continuous scenario planning matters

Continuous scenario planning becomes strategically important when it gives leaders a way to compare options side by side, test trade-offs before they commit, expose bottlenecks early, map dependency cascades and continuously recalculate what changes when budgets, priorities or constraints shift. That directly addresses many of the structural drivers identified above. It does not solve every reason projects fail. It does attack a large share of the root causes beneath them.

Seen this way, many of the familiar 25 reasons collapse into a smaller set of systemic failures. An unclear scope often results in a weak scenario definition. Poor requirements are often commitments made before constraints and dependencies were visible. Scope creep is often an unmanaged consequence. Poor communication often reflects fragmented planning logic, with business, finance and delivery working from different maps. Resource shortages are often hidden by overcommitment. Weak sponsorship often reflects weak evidence. Poor change control usually means the organization can record changes but cannot model impact. At the project level, teams can sometimes survive these problems through heroic effort. At the portfolio level, heroics stop working. Constraints win. Bottlenecks win. The question is whether leadership can see them early enough to respond intelligently.

PMI’s newer M.O.R.E. framework supports this shift. PMI argues that project outcomes improve materially when organizations manage perceptions, own success, relentlessly reassess and expand perspective. Two of those ideas matter especially here. Relentlessly reassess describes a discipline of continuous adjustment as conditions shift. Managing perceptions requires communicating value and risk in ways stakeholders can act on. That is remarkably close to what mature continuous scenario planning should do at scale.

Why the urgency is rising

The pressure on CIOs is increasing, not falling. EY’s 2026 CEO Outlook says leaders are pursuing growth and adaptability through bold AI transformation, with 2026 becoming a turning point as organizations move from pilots to scaled enterprise use. Accenture makes a similar point from a different angle, arguing that organizations that build strong AI foundations will be better positioned to reinvent, compete and achieve new levels of performance. Those are reasonable claims, but they do not reduce the need for disciplined planning. Faster change increases the premium on a planning system that can calculate consequences quickly and credibly. AI can accelerate analysis, summarize scenarios and improve executive access to planning insight. It cannot replace the need to govern trade offs across budgets, capacity, architecture, timing and risk. In fact, AI is only trustworthy in this context when it is tightly coupled to mathematically sound planning data, explicit constraints, dependency logic and algorithmic calculations. Otherwise, it risks producing plausible but unsupported answers.

What CIOs should demand

For CIOs, this leads to a more useful conclusion than simply restating the 25 reasons projects fail. Large programs usually fail because the enterprise cannot see and govern the interaction of those reasons in time. A modern control system for change, therefore, needs at least six capabilities: A unified planning model across priorities, budgets and capacity; side-by-side scenario comparison; interdependency mapping; early visibility into bottlenecks; continuous recalculation as conditions shift; and executive-facing summaries that turn data into decisions. Those are the capabilities that make continuous scenario planning strategically important. The question is no longer whether planning happens. It already does. The real question is whether planning remains static, fragmented and largely narrative, or whether it becomes dynamic, scenario-based and decision-grade.

That is the real fix hidden beneath the 25 symptoms.

This article is published as part of the Foundry Expert Contributor Network.
Want to join?

Technological sovereignty is often debated in terms of jurisdiction, compliance, or vendor origin. All of that is important, but it leaves out the important issue of retaining critical knowledge, which directly impacts the CIO.

Case in point, British bank TSB undertook a critical platform migration in 2018. The operation relied on a structure that, on paper, had guarantees of a validated provider, testing, and formal program governance.

Once the migration was complete, the new platform began experiencing technical difficulties, resulting in a significant disruption to branch, telephone, online, and mobile banking services, affecting a large portion of its 5.2 million customers. The extent of the matter was so complex, key problems weren’t resolved until the end of the year.

The crisis also had a significant economic impact. Banco Sabadell, which acquired TSB in 2015, had to absorb losses exceeding €200 million, and four years later, in December 2022, British regulators imposed a combined fine of nearly £49 million on the bank for failures in operational risk management, governance, and outsourcing supervision related to the migration. Then, in April 2023, the Prudential Regulation Authority, the Bank of England’s prudential supervisor, personally fined the then CIO £81,620 for failing to take reasonable steps to ensure adequate supervision.

The lesson from this case isn’t that a large migration can go wrong. Every CIO knows that can happen. But TSB didn’t have the capacity to govern and question critical vendor dependency.

The constant of knowledge

When we talk about technological dependence, we usually think of market concentration, long-term contracts, proprietary formats, migration difficulties, or negotiating power with the vendor. All of that exists and will continue to be important. But knowledge dependence is another form that comes up in the conversation and has a greater impact on the CIO’s day-to-day work.

This occurs when the organization doesn’t retain enough internal knowledge to discuss the technology, or subject it to serious scrutiny.

The TSB case was a clear example. The oversight of a critical department relied too heavily on unquestioned supplier guarantees. In other words, there was insufficient internal capacity to rigorously govern the outsourcing relationship.

With this example, the meaning of lock-in changes. It no longer manifests itself when migration becomes prohibitive or when an architecture becomes unchangeable. It begins earlier, when the company is operating its technology but can no longer reliably evaluate it.

In fact, this dependency isn’t easy to perceive because it coexists with a sense of reasonable operation. The services are available and the providers respond, and yet risks are being taken.

On the other hand, it forces a broader definition of sovereignty. The issue goes beyond where the data resides, under what jurisdiction a provider operates, and what degree of regulatory exposure a platform introduces.

Another question is how much critical knowledge does the company retain about what underpins its operations. From this perspective, maintaining sovereignty doesn’t equate to reviewing ownership of the technology or internalizing its implementation, thus preventing reducing the conversation to a legal or geopolitical debate.

Hidden knowledge dependencies

The common mistake when discussing tech dependence is to focus solely on the noisiest areas like cloud computing, AI, large platforms, and data storage. When discussing knowledge dependence, it’s essential, but not always easy, to look inward.

One area to consider is the architecture. Even if systems are functioning, it may become increasingly difficult to answer basic questions, like why the environment is designed this way, which parts are replaceable, or what changing a critical layer would entail. If this is the case, it’s a sign of dependency.

Another aspect is the operation. Outsourcing execution can make perfect sense, but problems arise when understanding is also outsourced. That is, when the internal team needs to go externally to make decisions or solve problems.

Dependency can also be hidden within the complexity of technological layers. In other words, it doesn’t necessarily have to be directly linked to a large platform, but to the set of integrations and connectors surrounding it, or a partner ecosystem that’s become a tangled mess. If no one understands the complete picture, dependency exists.

The knowledge CIOs can’t afford to lose

All of this shifts the focus to the specific responsibility of knowledge. Not all capabilities carry the same weight or have the same strategic value. But there’s a decisive threshold, the moment when the organization no longer sufficiently understands a dependency to be able to manage it. From that point on, the risk extends beyond the operation itself. The quality of decisions deteriorates, the CIO’s ability to discuss risks or costs diminishes, and many aspects end up being accepted without clear rationale.  

If it isn’t detected in time, there’s a risk of reaching a point of no return, where control of the technological roadmap is lost.

The debate for the CIO

The solution isn’t necessarily to distrust suppliers or outsourcing on principle. There’s a more subtle and demanding issue for the CIO of clearly deciding what knowledge can and can’t be shared externally. So the debate on sovereignty needs to become more pragmatic and more linked to the company’s actual capacity to understand what it depends on, and to change course when necessary.

In an environment of complex platforms, encapsulated services, and outsourced intelligence, preserving decision-making capacity will be an indisputable condition for technological autonomy.