AI Agents: From Experiment to Operating Assumption

See also (wiki): assistive-to-agentic-shift · agentic-ai-governance

Executive Summary

• 83% of global executives expect AI agents to improve process efficiency by 2026, per IBM’s largest agent-specific survey to date (n=2,900, Oxford Economics partnership, June 2025) — not a marginal bet, but near-universal expectation
• AI investment is on track to double as a share of IT budgets: from 12% of IT spend in 2024 to a projected 20% by 2026, with 64% of current AI budgets now allocated to core business functions rather than experiments
• The companies capturing real revenue from AI share one pattern: 52% revenue growth among self-identified “AI-first” companies (25% of the n=2,500 sample), compared to the broad population — but this is self-reported correlation, not a controlled experiment
• The three barriers blocking the remaining 75% from AI-first status: data readiness (49%), trust deficits (46%), and skills gaps (42%) — none of which resolve by purchasing more software
• ⚠️ TIER 2 SOURCE: Published June 2025, pre-current model generation. Agent capabilities have advanced materially since this fieldwork; adoption figures may understate current deployment velocity

The Expectation Gap That Matters

The more useful data point in this IBM survey is not what executives plan — it is the gap between what they expect and what the deployment data shows.

83% of the n=400 C-suite pulse respondents expect agents to improve process efficiency by 2026. That forecast was made in mid-2025. Cross-reference it against the current deployment reality from IBM’s own broader corpus: the global base of AI-enabled workflows stood at 3% at fieldwork time, with a projected jump to 25% by end of 2025. That gap — between 3% of workflows running on agents and 83% expecting measurable efficiency gains — is the central planning challenge for any CIO or COO reading this today.

The expectation is not wrong. The timeline may be.

Where the Budget Is Moving

AI investment is no longer a line item on an innovation budget. IBM’s n=2,500 survey finds 64% of AI budgets now allocated to core business functions — procurement, finance, customer operations, supply chain — rather than to skunkworks or R&D experiments. That shift in budget composition is the leading indicator, not a lagging one.

The trajectory: AI accounted for 12% of IT spending in 2024. IBM’s respondent base projects 20% by 2026. For a mid-market company spending $10M annually on IT infrastructure and software, that math implies roughly $800,000 moving from traditional tooling into AI tooling or AI-enabled workflow redesign over two years — before headcount or productivity effects are counted.

The 71% who believe agents will autonomously adapt to changing workflows are also making a prediction about architecture, not just deployment. Adaptive agents that self-adjust to workflow changes represent a meaningfully different engineering commitment than static automation. Most organizations buying agents today are not yet buying adaptive ones.

The AI-First Companies: Real Signal, Real Caveats

The headline that travels farthest from this IBM study is the 52% revenue growth figure attributed to “AI-first” companies. It is worth examining carefully.

The 52% revenue growth comes from a self-identified subgroup — roughly 25% of the n=2,500 sample who meet IBM’s definition of “AI-first.” This is not a randomized comparison. Companies that describe themselves as AI-first may share other characteristics (industry, size, prior technology investment, leadership quality) that independently predict revenue growth. IBM’s commercial interest in demonstrating that AI-first = revenue growth is significant. Oxford Economics’ partnership adds methodological rigor to the sample design, but not to the causal inference.

That said, the directional signal is consistent with independent evidence. BCG’s “Build for the Future” study (n=1,250, September 2025) found that companies they classify as “future-built” — a different definition, but a similar concept — achieve 1.7x revenue growth, 3.6x total shareholder return, and 1.6x EBIT margin versus the peer laggard group. McKinsey’s State of Organizations 2026 (n=10,000+) corroborates the pattern at the organizational level. The 52% figure should not be cited as a precise outcome, but as a directionally consistent data point in a body of evidence pointing the same direction.

The Three Barriers That Software Doesn’t Fix

Executives in the IBM survey named three adoption barriers ahead of everything else: data concerns (49%), trust deficits (46%), and skills gaps (42%). These are not feature requests. They are organizational conditions that must be built before agents can operate reliably.

Data readiness means structured, accessible, governed data — the prerequisite for any agent that touches a business process. Organizations without a data layer have a sequencing problem: you cannot profitably deploy agents into a data environment that was not designed for machine consumption.

Trust deficits divide into two subcategories: trust in AI accuracy (will the agent make mistakes that cost money?) and trust in AI governance (does the organization have visibility into what agents are doing and why?). The IBM IBV’s separate agentic cybersecurity analysis (n=1,000+, also 2025) shows that only 47% of enterprise agents are actively monitored. Trust without monitoring is hope, not governance.

Skills gaps are not primarily technical. IBM’s broader talent research shows the most acute shortage is in prompt engineering, agent workflow design, and AI governance — not in model development. Organizations that read “skills gap” as “hire more data scientists” will close the wrong gap.

Key Data Points

Source credibility rating: MEDIUM. Oxford Economics partnership and n=2,900 combined sample are strong. IBM has direct commercial interest in AI consulting and watsonx platform adoption. The 52% revenue growth figure is self-reported, correlation-only, and should not be treated as a causal estimate. Corroborating evidence (BCG “Build for the Future,” McKinsey State of Organizations 2026) is directionally consistent but independently derived.

What This Means for Your Organization

The IBM data captures a specific organizational moment: the transition from “evaluating whether to invest in agents” to “deciding how fast to go.” The 83% efficiency expectation is less a finding than a forcing function — when your board, your peers, and your vendors all expect agents to be operational by 2026, the decision to wait requires explicit justification.

Three decisions follow from this data. First, the budget shift to core business functions (64%) signals where the highest-confidence use cases live: accounts payable, customer service, procurement, and any workflow that is well-documented, data-rich, and measured by throughput. Those are the right starting points. Second, the 3% → 25% workflow projection creates an internal credibility test: any organization not tracking which of its core workflows are candidates for agent integration is flying blind when the budget conversation comes. Third, the barrier hierarchy (data → trust → skills, in that order) is a sequencing prescription. Organizations that invest in skills first, without first resolving data architecture and governance, will spend the budget and miss the outcome.

The 52% revenue growth figure should not drive a capital allocation decision on its own — the methodology does not support that weight. But it reinforces a pattern that now has consistent support across IBM, BCG, McKinsey, and Bain: the gap between organizations that restructure workflows around AI versus those that add AI to existing workflows is widening faster than most board-level AI strategies assume.

If any of this raises questions about where your organization sits in that distribution, the conversation is worth having — brandon@brandonsneider.com.

Sources

Brandon Sneider | brandon@brandonsneider.com April 2026