4 recs for CIOs as they implement agentic AI

Last year, in 5 ways CIOs can help gen AI achieve its lightbulb moment, I wrote about how we were in more of a gaslamp moment, with widespread experiments and accidents, as opposed to a true lightbulb moment. I noted then that despite the current level of hype and mainstream adoption, gen AI still needed to experience an expectation reset before embarking on a path to peak productivity.

Today, the hype is even more exaggerated with agentic AI as we read daily proclamations about the end of traditional consulting and programming. And we’re told we should be implementing projects in minutes or hours, not days or weeks. Seasoned influencers, to their credit, add caveats such as “for prototypes, not production.” 

It’s certain that AI’s lightbulb moment will come, as well as that for agentic AI, as firms move from digital-first to AI-first strategies. But given the technology’s current position in the hype cycle, we still need to pass through, and emerge from, the trough of disillusionment.

We’re seeing signs of stress, such as in Gartner’s 2025 AI Implementation Survey, where 87% of Indian global competency center leaders have made public statements about their gen AI initiatives, yet only 23% have deployed solutions that materially impact business outcomes. And KPMG’s Voice of the CIO paper says that issues and angst are mounting among CIOs as they’re goaded into accepting tech provider forays into gen AI functionality.

So to get a sense of how CIOs are approaching their agentic AI implementations, a number of tech leaders describe here how they’re proceeding, whether they’re taking it fast or slow, and their thoughts on a reset of expectations as we see more real-world implementations and issues coming to light. Based on these conversations, here are four recommendations for CIOs on how to approach an agentic AI implementation.

Decide how fast you should go, not can go

According to Dan Garcia, CISO at software developer EDB, he and his team recognize that different agentic AI use cases require varying degrees of pacing. They move fast in areas where the business ROI is clear, where there’s mature data infrastructure, and where governance allows. And they have to be more deliberate in areas where automation, hallucinations, safety, or misuse of data can impact the value proposition they’re seeking. “We see these common threads among our customers,” he says. “Delivering agentic AI promises requires sovereign infrastructure that provides control of your data, logic, and business outcomes. So the question isn’t how fast you can go, it’s how fast you should go, and where you still need a human in the loop to maintain trust and accountability.”

A hybrid mindset is becoming the norm, according to Anirudh Narayan, chief growth officer at AI agent platform Lyzr.ai. “CIOs are deliberately choosing a ‘fast in pilots, slow in production’ mindset,” he says. “They’re rapidly experimenting with agentic AI in isolated workflows to capture quick wins but scaling enterprise-wide only after clear success metrics are met, particularly around security, observability, and human-in-the-loop validations. The decision factors center on regulatory risk, quality of internal data, and the availability of skilled operators who can maintain these agents post-deployment.”

So if you’re integrating an AI vendor solution into your own custom applications, you’ll need a skilled team to manage and maintain these agents after deployment.

Let use cases and enterprise complexity guide you

Marcus Murph, head of technology consulting at KPMG, is seeing a hybrid approach among clients where most organizations look to balance agility with caution. Many are starting in lower-risk domains where they can show early ROI and build the foundations needed to scale agentic AI over time. But the more telling pattern, he says, is based on enterprise complexity. “Large, established companies are moving at a more measured pace, often because they’re navigating decades of legacy systems, deep-rooted workflows, and tech debt,” he adds. “On the flip side, smaller organizations or newer players who aren’t weighed down by those constraints are moving faster and with more flexibility.”

To manage risk and complexity, CIOs can also explore other foundation models such as SLMs instead of LLMs. Here, use cases can be the guide and in many instances SLMs are better-equipped to deliver business-specific AI apps.

Design for integration and scale

The tech leaders I spoke with all agree that a reset is likely. According to Murph, just like we saw with gen AI, the hype always gets ahead of the reality, and you can already feel the friction between ambition and execution with agents. “The good news is some clients are already preparing,” he says. “They’re not just building agents, they’re building the scaffolding around them. That means putting the right guardrails in place, managing stakeholder expectations, and designing for integration and scale, not just proof of concept.”

He also sees the enterprises that will treat AI as primarily a business problem, as opposed to a tech problem. Success will require substantial rewiring of business process and re-shaping of the workforce, in addition to an implementation of technology.

EDB’s Garcia also thinks a reset isn’t just coming, but is predictable given how fast businesses want to move while needing to balance management, scalability, and deliver value on investments. “We’re at the peak of inflated expectations, and CIOs are holding the mop for the fallout of overpromises,” he says. “The proactive leaders we’re working with are insulating their teams from hype by running controlled pilot environments with guardrails and budget thresholds, and are continuously learning as they iterate.”

Seize the chance to step up

Since CIOs have to keep the lights on as well as innovate as part of their day job, they’re in a strong position to lead the charge around AI and agentic AI, and take a well-measured approach regardless of its stage in the hype cycle.

Data from KPMG’s recent AI Pulse Survey shows that leadership of AI is shifting from CEOs to CIOs, where in Q1 this year, 86% of CIOs lead AI-related initiatives compared to just 8% of CEOs. This is a significant shift from the same period last year when the numbers were 31% and 34%, respectively.

CIOs are working closely with CAIOs as well, and helping them pace appropriately. “There’s a new sense of responsibility emerging,” says Garcia. “CIOs are managing CAIOs who are eager to move fast, but often don’t give enough attention to data readiness, legal risk, or long-term operational complexity. It’s not about slowing things down — it’s about pacing with purpose.”

Murph sees this as a moment for CIOs and CAIOs to work as copilots. “The CIO knows the tech stack; the CAIO knows business transformation,” he says. “Neither can do this alone. What we’re advising is move fast and smart. That means aligning on what’s ready to be piloted now, what needs guardrails, and where the tech still has room to mature. If you don’t have that shared roadmap between the CIO and CAIO, that’s when velocity turns into volatility.”

AI’s lightbulb moment will come, which is inclusive of gen AI, agentic AI, and more. To navigate getting there, CIOs and CAIOs can take a ‘pacing with purpose’ approach, let enterprise use cases and complexity guide them, design for integration and scale, and continually strive to strike the right balance between risk and reward.