A Tale of Two AI Cities, Or Is It?
AI works. Strategy is the harder problem.
Why AI strategy is shifting from proving what works to deciding what matters
Why AI may be enabled by technology—but ultimately judged as a business decision
And why leadership alignment, not agents, determines what comes next
“Efficiency gains are the primary unlock from AI agents.”
Anthropic, 2026 State of AI Agents Report“Executives expect AI to increase productivity by 42% by 2030, and 67% expect to have captured most of those productivity gains by then.”
IBM Institute for Business Value, The Enterprise in 2030
I came to this topic intending to write about what are still widely described as the hard parts of AI: access to quality data, integration with legacy systems, and the challenge of connecting AI to the applications that actually run the business.
But reading these two reports side by side,, along with my earlier Substack piece on IBM’s Enterprise in 2030, the contrast between them pulled me in a different direction. Not because either report is wrong, but because together they raise a more fundamental question:
If IT proves AI works, but the business decides where AI matters next, who owns AI strategy?
Two perspectives, not two verdicts
At face value, the difference between these reports is straightforward.
Anthropic’s State of AI Agents reflects a near‑term, IT‑led perspective. Based on a survey of 500 U.S. technical leaders, it highlights where AI and agentic systems are already delivering value today: software development, workflow automation, data analysis, and productivity gains. The emphasis is pragmatic, agents work, efficiency improves, and those gains are expected to continue compounding over the next 12 months.
IBM’s Enterprise in 2030, by contrast, reflects a longer‑horizon, business‑led view. Drawing on line‑of‑business and senior enterprise executives, the report assumes many of those efficiency gains will be largely realized by the end of the decade. As a result, IBM’s respondents anticipate a shift in focus, from productivity improvements toward product, service, and business‑model innovation.
These are not competing truths. They are views from different seats, looking at different points on the timeline.
What “efficiency first” really means
It is tempting to interpret today’s AI success stories as conservative or incremental. I think that misses the point.
Efficiency‑oriented use cases, coding assistance, automation of routine tasks, improved reporting, are not dominant because they are unambitious. They succeed because they align with how organizations currently operate:
Ownership is clear
Risk is contained
Value can be demonstrated quickly
Human judgment remains firmly in the loop
Anthropic’s report is explicit about this focus, and it is a rational one. These are the use cases that fit today’s organizational, cultural, and accountability models.
The looming handoff to the business
IBM’s research points to what comes next. If productivity gains are largely captured by the end of the decade, as executives expect, then efficiency stops being the differentiator. Innovation becomes the mandate.
This creates a tension that neither report directly addresses:
If IT leaders are driving near‑term AI value through efficiency and automation, but line‑of‑business leaders are expecting AI to fuel growth, differentiation, and new offerings, who ultimately owns the AI agenda?
History suggests an answer.
We have seen this pattern before. CRM, sales enablement, and marketing platforms did not gain momentum because IT validated them first. They gained traction because business teams, particularly marketing and sales, were not getting the outcomes they needed from existing IT‑delivered solutions. Value creation pulled those technologies into the business long before governance caught up.
AI shows signs of following a similar arc.
A brief note on data and integration
Both reports acknowledge familiar constraints: access to quality data and integration with existing systems and applications.
These remain real challenges, but they are better understood as organizational issues as much as technical ones—questions of ownership, incentives, and priorities. AI exposes them quickly, but it does not resolve them automatically.
Taken together, these two outlooks offer a useful lens.
Anthropic shows us where AI is delivering value now, largely through IT‑led efficiency gains. IBM highlights where business leaders expect AI to matter next, once those gains are absorbed.
The gap between the two is not a failure of technology. It is a signal that AI adoption is approaching a familiar transition point—from proving feasibility to redefining business value.
Where the real work lies
AI will continue to improve. Agents will become more capable. Automation will deepen.
Meaningful transformation will remain gated by human decisions, particularly those made by line‑of‑business leaders who define success, own outcomes, and bear risk.
Until IT and the business are aligned on that transition, efficiency will dominate not because AI lacks imagination, but because organizations still rely on humans to decide what matters.
That, more than any model or agent architecture, will determine where AI creates lasting value.
And as AI moves from proving what it can do to deciding what it should do, the open question remains: who is prepared, and empowered, to make that call?
Sources of information for this article:
Anthropic’s How enterprises are building AI agents in 2026
IBM’s The Enterprise in 2030
My Substack Article The Enterprise of 2030 isn’t AI-first. It’s intent-first.