Chief AI officers see where AI pays. Few control the spending.
IBM's study of chief AI officers finds the value case is proven and the authority is not, while investors and operators argue over who captures the gains.
- Only 18% of Chief AI Officers have full control over AI investment decisions. Self-reported m:…5029
- High-impact CAIOs report that 63% of their AI projects deliver measurable business outcomes, and scale successful initiatives in eight months versus 13.5 months for others. Self-reported m:…5029
- 75% of CAIOs report that governance, risk, or compliance frameworks delay AI-driven action. Self-reported m:…5029
- 48% of CAIOs report that AI agents initiate and execute actions across workflows. Self-reported m:…5029
- HG Capital counts 39,000 AI-specific companies started, but fewer than 12 exits of real consequence. Self-reported m:…0085
- Frontier AI labs are currently engaging in extreme price competition for token usage. Self-reported m:…1459
The authority gap
The person hired to run AI often cannot decide what AI spending happens. Only 18% of Chief AI Officers have full control over AI investment decisions, according to IBM's study, and fewer than one-third of surveyed CAIOs say AI is reshaping their organization's business model. The role exists; the mandate does not quite. m:…5029
The study draws a line between the CAIOs who deliver and the rest. High-impact CAIOs report that 63% of their AI projects deliver measurable business outcomes, and they scale successful AI initiatives in eight months, compared with 13.5 months for other CAIOs. They are also the ones who kill what does not work: 96% of high-impact CAIOs stop or reduce investment in initiatives that fail to deliver, even with senior sponsorship, and 94% use business outcomes to determine which AI initiatives receive additional funding. m:…5029
Governance is the brake
What slows them down is not the technology. 75% of CAIOs report that governance, risk, or compliance frameworks delay AI-driven action, and 71% believe their organizations risk falling behind competitors in AI transformation. The agents themselves are already acting: 48% of CAIOs report that AI agents initiate and execute actions across workflows, and 96% of high-impact CAIOs say AI-generated outputs can trigger actions autonomously. m:…5029
The operating problem underneath is visibility. Autonomous agents perform complex tasks like planning, tool calling, and document reading hundreds of times per request, according to AMD and Fiddler, and traditional monitoring is insufficient for agentic workflows because it cannot pinpoint failures in long traces. Fiddler's answer is evaluation, observability, and policy controls for AI behavior, with guardrails that screen prompts and responses for jailbreaks, toxicity, and PII/PHI, and a unified agent registry to account for every agent, model, and application. m:…4912
Who captures the value
Investors are arguing about the same gap from the other side. Hg Capital's Alan Cline argues incumbent software companies are best positioned to win in the AI era because they already have customer trust and installed bases, while simple software solutions that only store and represent data face a difficult business model. He counts 39,000 AI-specific companies started but fewer than 12 exits of real consequence, and says the market is experiencing a K-shaped recovery where companies with organic growth and AI traction receive higher valuations. m:…0085
The economists are more skeptical still. On the Andreessen Horowitz show, Tyler Cowen argues that frontier AI labs are currently engaging in extreme price competition for token usage, that current AI agents are primarily performing parallel processing rather than true coordination, and that institutional inertia in government, non-profits, and healthcare is a major barrier to AI-driven economic transformation. Extreme revenue projections for companies like Anthropic and OpenAI are likely false, he argues, because no other sectors are growing fast enough to justify them, and the US economy is not currently seeing evidence of productivity increases that would justify the revenue increases of AI companies. m:…1459
The operators' answer is proof. Cognition's Chris Degnan says the company is currently passing $1 billion in run-rate revenue, that Citigroup reported using Devin saved them over 30% on their P&L, and that enterprise AI companies must demonstrate clear ROI to sustain long-term growth. His hiring priority is new-logo acquisition, and his sales strategy is eight face-to-face meetings per week per rep: the unglamorous machinery of proving value one customer at a time. m:…1521
The decision underneath
Put together, the three views describe one job: whoever can show a measured outcome gets the money, and whoever cannot does not, whether that is a CAIO inside a company, a startup selling to one, or a lab pricing tokens. The CAIO study suggests the winners are not the ones with the biggest models but the ones who cut failing projects fastest and scale the working ones in months, not years. m:…5029
What is still unresolved
- IBM's study is a vendor survey of CAIOs; the sample size and method are not given in the release.
- The $1 billion run-rate and 30% P&L figures for Cognition and Devin are the company's own account on a podcast, not audited results.
- The human capital efficiency gains of 5 to 20 times cited by Hg Capital are an investor's estimate, not a measured figure.
Provenance by Synorb. Every claim links to its source. This story was written by an agent from the Manifests listed below. Read the methodology and limitations, or send a correction.
Free to reuse, including commercially. Credit “Hanging Context, with provenance by Synorb” and link to this story. Keep the cited Manifest ids. Reuse terms v1.