Opinions expressed by Entrepreneur contributors are their own.
Key Takeaways
- AI agents are becoming customers, but companies rarely disclose how much revenue they drive.
- Agent-driven revenue can shift overnight when foundational models retrain or change defaults.
- Investors need new disclosure standards to measure concentration risk in the agent economy.
Public market investors are flying blind into the volatility trap of the agent economy. When a SaaS company reports 40% revenue growth, there’s no way to know if that growth comes from durable customer relationships or AI agents that could disappear with the next model update.
The accounting hasn’t caught up to the reality of who’s actually buying.
In a recent article, I argued that traditional metrics like customer acquisition cost and lifetime value break down when your customers are AI agents that have no loyalty or memory. I proposed “agent penetration rate” as a better framework. But knowing how to measure something and actually requiring companies to disclose the results are different problems.
Without that transparency, AI-driven revenue creates concentration risk that just doesn’t exist with human customers. Companies that segment this exposure give investors essential information. Those who don’t should face pressure until they join a new disclosure standard.
In fact, we might eventually need modifications to GAAP.
The volatility trap
Traditionally, companies could lose their biggest customer and take a revenue hit, but this was an expected and measurable phenomenon. Measures like retention and churn tell us in aggregate what we should expect. AI-driven revenue concentrates differently.
When coding agents select authentication providers or payment processors, they’re following patterns learned during training by a handful of foundational models. For example, if OpenAI retrains GPT and shifts toward a competitor’s API, every application generated after that update defaults to the new choice. Revenue can shift substantially toward whoever appears most frequently in the newest training data.
In other words, the winner takes all.
It’s an issue that goes beyond developer tools. When a procurement AI used by large enterprises re-trains on updated vendor data, it can shift overnight from selecting one supplier to another. There’s no loyalty buffer, no switching cost, no relationship to preserve. The selection follows training data patterns rather than economic indicators like recessions or recoveries.
This makes stress-testing nearly impossible using conventional models. Investors evaluating a company with significant agent-driven revenue can’t rely on historical recession performance or standard sensitivity analysis. But right now, there’s no way to measure exposure because companies don’t disclose it, and oftentimes don’t measure it.
The transparency gap
Many companies cannot even tell you the disparity between human versus agent revenue because they aren’t tracking it. Segmented reporting would change this.
I suggest companies break out revenue by source: human-driven purchases, agent-driven integrations, and hybrid transactions (where humans approve agent recommendations). For agent-driven revenue, they should disclose concentration across foundational models and estimate what percentage depends on current training data distributions versus locked-in enterprise contracts.
All of this will require analyzing integration patterns, tracking the transactions that originate from AI tooling, and estimating exposure to foundational models. Yes, it’s complex, and some resistance is to be expected.
Management teams may cite the high technical cost of tracking agent patterns. It might also just be impossible to measure unless the agentic “user” self-discloses. Companies might consider treating it like measuring user acquisition funnels or marketing attribution today. When users sign up, websites sometimes ask the user where they heard about their service. It’s voluntary disclosure, but it’s helpful for properly allocating marketing spend.
Treating all revenue as equivalent when the underlying volatility profile is so divergent misrepresents risk to shareholders. It can also leave value on the table. If a company is not performing well with agentic users, it’s worth asking why. The company’s generative engine optimization (GEO) may need further investment.
Investor implications
For public market investors, the practical question is how to price the current opacity. One approach is to apply industry comparables if a software company cannot or will not articulate what percentage of revenue comes from agent-driven integrations.
The comparable could result in a discount or a premium. A discount reflects two risks — concentration exposure and the possibility that management doesn’t understand their own customer base well enough to measure it. A premium might be applied if it’s clear that a company has some unfair advantage in agentic user acquisition or if their customers are retentive for some other qualitative reason.
Vertically integrated software remains the exception. When a company sells directly to a single institutional customer — a hospital system, a government agency, or a large bureaucratic enterprise — the end user is still human. Those customers don’t churn based on LLM training cycles.
But even vertically integrated businesses may have products that include agent-driven components. A healthcare software platform, for example, might include AI assistants for physicians. Investors need to understand where the agents sit in the value chain and how much influence they have over which features get used. This is a new form of due diligence that will only grow in relevance.
Making disclosure standard
The gap between what companies report and what investors need to know is widening with agentic AI coming on stream. It is redefining competitive advantage, and the question for investors is how to properly price an asset when the user base is poorly understood.
Waiting for FASB or the next market correction is a passive choice that creates opportunities for savvy investors. I believe that more accurate reporting is better for our financial markets, because it will lead to more efficient capital allocation to the companies that deeply understand their user base — agentic or otherwise.
Key Takeaways
- AI agents are becoming customers, but companies rarely disclose how much revenue they drive.
- Agent-driven revenue can shift overnight when foundational models retrain or change defaults.
- Investors need new disclosure standards to measure concentration risk in the agent economy.
Public market investors are flying blind into the volatility trap of the agent economy. When a SaaS company reports 40% revenue growth, there’s no way to know if that growth comes from durable customer relationships or AI agents that could disappear with the next model update.
The accounting hasn’t caught up to the reality of who’s actually buying.
