OpenAI and Anthropic's PE Push Could Make Agent Washing Worse

The obvious read is that private equity just became AI's new distribution channel.

The more useful question is whether that channel can carry real deployment before the operating base is ready.

In early May, Axios reported that OpenAI and Anthropic were teaming up with private equity firms on multibillion-dollar ventures to push AI tools into midsized companies. OpenAI announced the OpenAI Deployment Company, backed by more than $4 billion of initial investment and led by TPG, with Advent, Bain Capital, and Brookfield as co-lead founding partners. Anthropic announced a new enterprise AI services company with Blackstone, Hellman & Friedman, and Goldman Sachs.

That may become an important route for AI into operating businesses. It also may not stick in the form announced this month. Announcements are not adoption, and adoption is not proof of value.

But the operating pressure is already here.

This is not just a channel story

Private equity is attractive because it owns companies with repeatable operating problems. One sponsor can turn a working pattern in one portfolio company into a mandate across ten more.

That is the real shift: AI moves from side experiment to board-level operating pressure.

OpenAI and Anthropic are both describing the same gap. OpenAI says the Deployment Company will embed forward deployed engineers and connect models to customer data, tools, controls, and business processes. Anthropic says mid-sized companies can benefit from AI but often lack the in-house resources to build and run frontier deployments.

If model access were enough, these partnerships would not need to exist. The hard part is turning a general capability into a reliable operating workflow: understanding the process, edge cases, incentives, data, and accountability well enough that AI changes how work happens instead of decorating the old operating model.

OpenAI and Anthropic may approach this work with the seriousness it deserves. The risk is that the channel creates more pressure than the operating base can absorb. If the mandate outruns workflow readiness, data quality, and accountability, PE-backed companies can end up buying agentic-sounding systems before the business is ready to use them. That is where agent washing becomes a real operating risk.

The base is still uneven

FTI Consulting's 2026 Private Equity AI Radar shows the tension clearly. PE funds are not ignoring AI. FTI says 95% of funds report AI initiatives meeting or exceeding their original business case criteria, and revenue acceleration is the top cited priority.

But only 36% of portfolio companies report AI deployed across use cases, and only 7% report enterprise-scale deployment.

Those numbers can both be true. A project can meet its initial business case and still fall short of enterprise-scale operating adoption.

In our last piece, we looked at a similar pattern in the broader enterprise market: companies can accumulate tools without building a knowledge system that actually makes the business smarter. The press release says "AI." The org chart says "AI." But the company still does not know how to turn all that activity into better decisions.

The PE version has the same shape: real interest, real pressure, uneven deployment.

So the question is not "will PE-backed companies buy AI?" They will.

The question is whether those deployments help people do better work and create operating knowledge that compounds, or whether they create a new layer of AI-labeled workflow theater.

Gartner's warning labels

This is where Gartner's May 20 warning matters. Gartner was writing specifically about supply chain planning technology, so it should not be stretched too far. But the buyer-risk applies well beyond supply chain.

Gartner's point was that a lot of current "agentic" capability is still query interpretation, recommendations, and conversational support. Useful, maybe. But that is not the same thing as a system that generates plans, chooses the best plan, and executes without human intervention. Is the system assisting a human, recommending a next step, routing an exception, changing a replenishment parameter, or approving an invoice? Some of those are human-in-the-loop workflows. Some hand execution authority to the system. For most PE-backed companies, that fully autonomous version should not be the starting point.

"Agentic" is not enough information.

The buyer question is: what is this system actually helping someone do, what data supports that work, and who is accountable when it is wrong?

Gartner made the same point from the worker side two weeks earlier in a broader warning about autonomous business and AI layoffs: the returns came less from cutting people than from investing in the skills, roles, and operating models that let people guide and scale autonomous systems.

The data problem is more basic, and more pervasive, than people want it to be

Before the deployment team arrives, the boring questions matter: does the data exist, is it usable, and does everyone trust the same source of truth?

That last question is the one that gets skipped. A lot of companies have messier data than their AI strategy assumes. Different teams may be looking at different slices of the business and using those slices to argue different conclusions. Both can claim the data is on their side when there is no concrete source of truth.

AI does not magically resolve that disagreement. It can amplify it. It can make the output look polished while the underlying business is still arguing with itself.

That is why the first use case matters so much. A good first deployment should be narrow enough to inspect, measurable enough to prove, and low-risk enough that a human can stay accountable while the team learns where the system helps and where it breaks. In some companies, the first step is the foundational data, governance, and institutional knowledge work required to make the workflow usable.

The goal is not to buy autonomy on day one. It is to prove useful judgment in one workflow, help the people in that workflow get better at their jobs, and build the data, governance, and institutional knowledge needed to scale it.

The mandate is already moving

That pressure is moving through capital structures and boardrooms into operating teams.

Done well, a deployment team paired with a serious operator can make one workflow smarter and help that knowledge compound. Done poorly, "agentic" becomes boardroom shorthand for skipping the work.

If you are a PE-backed operator and the AI mandate shows up next quarter, the right response is not resistance or blind adoption. The standard is simpler: can this system improve a specific workflow, from trusted data, with a human still accountable for the decision?

That is the operating question behind the PE channel. Not whether the system sounds agentic. Whether it helps people make better decisions and proves it in the work.