One of the largest asset managers in the world said on October 24 that it can find no agentic AI payoff outside tech. None at all. Its chief economist wrote that AI “has delivered zero financial payoff to corporate America outside of tech.” He then published five charts to back the claim.
New here? I publish one agentic AI governance post every weekday. Subscribe to the blog and it lands in your inbox the moment it goes live.
The charts all tell one story. Revenue per worker is rising for the seven largest tech firms. It is flat for the other 493 in the S&P 500. Profit margins show the same split, too. Total factor productivity, the widest measure of output per unit of input, shows no AI lift at all. The note ended bluntly: “the AI story will be decided in the S&P 493, not the Mag 7.”
The economist offered a reason, and it is the right one. “AI does not understand accountability, trust, culture, unwritten rules, ethical trade-offs or fairness,” he wrote. Those are the things that drive revenue and margins in a real firm, after all. But he stopped one step short. Accountability and trust were never going to come from the model. They come from the harness built around it, and most companies never built one.
The spending is real and the payoff is not
The same note also shows where the money goes. Token spending, according to card data from Ramp cited in the note, sits with a small number of venture-backed firms. Most companies, the economist wrote, now treat tokens as “a cost center that weighs on margins.” So the firms that spend the most are the firms selling AI. The firms that should benefit are pulling back.
Anthropic’s own data points to the same agentic AI payoff gap. In its study of agent autonomy, published in February, about 73 percent of tool calls still had a human in the loop. Fewer than one percent of actions were one-way doors. In short, the agents that work today work because a person and a set of controls sit around them. The brains are not the scarce part.
Why the model cannot close the gap
A model is a calculator, the economist wrote, but business runs on judgment. That is true, as far as it goes. But judgment in a big company is not a feeling. It is a chain of decisions: what an agent may do, what it may touch, who signs off, what gets recorded. None of that lives inside the model weights. An architect can design all of it.
The agentic AI payoff shows up only when that chain exists. Without it, an agent produces an answer that nobody in risk, audit or finance can act on. The output is good and the output is useless, because nobody can defend it. So it sits in a pilot, the token bill keeps running, and the chart stays flat.
More on the Agentic AI Payoff
- The Core Definition: My plain-language answer to what agentic AI is and why intent in, outcomes out is the test.
- Govern the Harness: Why I argue that governance belongs on the harness, not the model.
- The Harness Term: Where “harness” came from and how each vendor uses it.
- Two Banks, One Lesson: Two lenders, one control design, and one very different outcome.
Two companies, two outcomes
Two cases from The Governed Loop show the same split inside single companies. For example, a large insurer bought enterprise copilots for several thousand staff in 2025. Usage was high and the vendor’s dashboards looked good. Still, the finance team could not find one closed claim that moved faster because of it. The agents had no written intent and no execution boundaries. There was no evidence trail a claims auditor would accept. Spending rose, but margin did not move.
A mid-size lender, meanwhile, took the other path. Before it let an agent touch a loan file, it wrote down the intent in business terms. It scoped the agent’s permissions to one stage of origination. Then it logged every tool call in a form its model risk team could read. The agent earned wider autonomy over three releases. Each step was signed off against the evidence. Cycle time fell, so the saving landed in a line the CFO could see.
The difference was not the model, because both companies used frontier models from the same two vendors. One company governed the loop around the model. The other one bought tokens. That is the agentic AI payoff gap at the scale of one balance sheet.
What I would fix first
My work in regulated loan origination taught me this the hard way, so I will say it plainly. The missing agentic AI payoff is a governance problem wearing a technology costume. The companies in the flat half of those charts already own a model that is good enough. What they lack is the six layers that turn a smart model into an outcome someone will sign.
Intent comes first: the outcome, written in business terms, with its limits. Capability follows, as a registry of what the agent may touch and what it may not. Enforcement sits under that, where the runtime checks execution boundaries and permission scoping, not a policy deck. The runtime itself is the harness that runs the loop and can stop it. Evidence is the audit trail a regulator can read without an engineer in the room. Autonomy comes last, earned one step at a time, never granted up front.
That is the whole argument of The Governed Loop. It is also the only version of the agentic AI payoff I have seen hold up in a regulated company. Instructions in, results out was IT. Intent in, outcomes out is agentic AI. The chart will move when the loop exists, not when the next model ships. Until then, the line stays flat.
Is your AI spend governed by a loop like this, or by a budget line?
Was this post useful? One click tells me what to write next.
Subscribe to my LinkedIn Newsletter
- Agentic AI Architect: control design for enterprise agents.
- Agentic AI Case Studies: deployment evidence, one teardown at a time.
- Agentic AI P&L: cost, payback, and risk in a CFO’s voice.
- Agentic AI SDLC: the lifecycle that builds the agents.
- Subscribe to the Blog: I publish a new post every weekday on one agentic AI governance question. Enter your email and each post arrives in your inbox the moment it goes live.
© Dr. Harish Kotadia, Ph.D., All Rights Reserved, 2026
Dr. Harish Kotadia, Ph.D., is an Enterprise AI Architect with 20+ years of IT consulting experience serving Fortune 100 clients, specializing in agentic AI systems built on Anthropic Claude, AWS Bedrock, and Google Vertex AI.
Disclaimer: This blog post is based on publicly available academic publications, vendor documentation, open standards, and news items from reputed media sources linked above. This post is intended for educational purposes, to help the enterprise agentic AI community build a shared vocabulary from public, authoritative sources.
Views and opinions expressed here are my own and do not represent those of any employer or client, past or present. The analysis presented is my independent interpretation of the published sources linked above and does not constitute legal, financial, or consulting advice of any kind.