Microsoft has yet to ignite enthusiasm for agentic AI

It’s the year of the agentic enterprise, according to Microsoft, which at its Ignite show this week positioned autonomous workflows and reasoning-driven automation as the almost-inevitable path that enterprises must eventually take.

But there’s reluctance, and resistance to the investments that will be required to take advantage of the new technology.

“Every organization will run on a distributed ecosystem of agents moving forward,” Microsoft’s EVP of Office and Copilot, Ryan Roslansky, told keynote attendees, while Microsoft’s president for business and industry Copilot, Charles Lamanna, followed up with, “Agents are showing up everywhere, embedded in every workflow. And these agents will be built by everyone in your organization.”

Agents, agents everywhere

Agents were indeed everywhere, mentioned over 250 times in the nearly-two-hour-long keynote hosted by Judson Althoff, CEO of Microsoft’s commercial business. He and other top executives from the company and its partners unveiled new products, layered with demos, that doubled down on the narrative that every enterprise must be an agentic enterprise to succeed.

At the center of that pitch were a handful of flagship offerings — Work IQ, Fabric IQ, Foundry IQ, and Agent 365 — which were positioned as tools that CIOs could adopt to leap barriers around context engineering and agent sprawl, forming the backbone of Microsoft’s argument that enterprises can finally move closer to a scalable agentic architecture.

However, Althoff and the other executives’ narrative is precisely what analysts identify as a sticking point for CIOs as they face pressure to scale agentic AI shifts from boardroom aspiration to operational reckoning.

“In 2024, CIOs were largely wrestling with experimentation, high costs, lack of skills, immature data estates, and pressure to deliver quick wins from generative and agentic AI. In 2025, the challenges have shifted from ‘can we do something interesting?’ to ‘how do we run this at scale, safely, and with predictable economics?’”, said David Linthicum, independent consultant and retired chief cloud strategy officer of Deloitte Consulting.

That shift in sentiment, according to Linthicum, has widened the gap between the optimism radiating from keynote stages and the realities inside most enterprises putting vendor-driven agentic hype ahead of organizational capability.

CIOs are still grappling with uneven data quality, fragmented system landscapes, and misaligned organizational structures, all of which remain the fundamental determinants of whether agentic AI can run reliably in production, Linthicum added.

And while Microsoft has talked about how it can automate some of the implementation of new tools like Fabric IQ, one the IQ offerings— building ontologies for semantic intelligence from existing data structures built for Power BI, for example —  Constellation Research principal analyst Michael Ni has highlighted the needed effort and functional collaboration, saying, “There is upfront work for IT. Ontologies don’t build themselves.“

Over-complication

Linthicum, for his part, criticizedthe branding of the IQ products as creating unnecessary complexity for CIOs.

“By splitting what is essentially a single architectural concept into multiple branded ‘IQ’ products, Microsoft may inadvertently confuse buyers and encourage over-purchasing,” he said. “The concern isn’t that these tools lack value individually, but that organizations might rush to activate every feature before establishing fundamental processes, without enterprises could end up with fragmented deployments that fail to deliver the promised efficiency and intelligence.”

The cloud strategy consultant was also skeptical of the upgraded Azure Copilot, saying that instead of helping reduce complexities around cloud management, it could introduce a few of its own.

The agentic updates in Windows 11 have been called out as a major pain point by some.

HFS Research CEO Phil Fersht highlighted the problems for CIOs and enterprise administrators concerned with controlling their desktops: “Enterprises cannot afford unpredictable agent behavior on employee devices. They want guardrails, clear opt-in controls, and predictable system behavior. Until Microsoft provides more transparent controls for enterprise environments, CIOs will be cautious about letting Windows-based agents act without human oversight,” he said.

However, for CIOs intent on embracing Microsoft’s new stack, analysts offer a few pointed recommendations.

Fersht said CIOs should work with business leaders to define the outcomes that agents are supposed to deliver, followed by rationalizing their data estates and putting up strong governance practices.

CIOs will likely need a whole separate function or department to monitor implementation of agentic AI, according to IDC’s group vice president of research for AI, data, and automation, Philip Carter. “CIOs should formalize AI Centers of Excellence (CoEs), appoint senior AI and data leadership,” he said, adding that these CoEs would be critical in developing enterprise-wide lifecycle policies covering agent design, safety, and ongoing oversight.

But CIOs should exercise caution while trying to pilot or scale these still-immature agentic capabilities, Carter warned. “The gap between demonstration capabilities and production-ready reliability can be significant, particularly given the industry’s general struggle with agentic AI ROI,” Carter said.