Agentic Factory

A capability, not a project

Vista Equity Partners says it is converting one portfolio company to generative AI every week, and by early 2026 had already moved more than thirty of the roughly ninety companies it owns.¹ The agentic factory is the standing internal capability behind that pace: a mechanism for moving an entire investment portfolio across a platform shift, applied serially to every company owned rather than performed once per deal. The idea's interest is that its builder claims to be running it for the second time, which turns an otherwise unfalsifiable AI promise into a claim with a prior.

The first version converted portfolio companies from on-premise software to the cloud starting around 2000. Enterprise software then ran on customer-owned hardware maintained by systems administrators. A repeatable conversion capability was built and applied across more businesses moving on-premise to cloud than almost any other institution, with a measured result of a two-and-a-half to three times economic rent pickup per conversion, sourced from eliminating the hardware refresh cycle and the systems-administrator layer to produce a structurally cheaper delivery model.¹

The second version, with infrastructure built around mid-2023, moves cloud companies to generative AI, already producing revenue at the companies converted so far, with another thirty to forty due within a quarter.¹

What the factory actually manufactures

The less obvious claim, and the one worth keeping, is that the factory's scarce output is not agents. Agents are cheap, and any portfolio company could build one. What is claimed to have been built instead is the ability to hit the precision enterprises require from probabilistic models inside real workflows: the capability to transform a business and create agentic products that work within existing workflows at the high precision enterprises demand.¹ Precision engineering against probabilistic models is expensive and largely portable between companies, so building it once and amortizing it across ninety balance sheets applies the logic of scale economies shared to an operating capability rather than to purchasing.

Three inputs make the factory work: a capability built centrally and amortized across the portfolio so no individual company pays to invent it; hyperscaler partnerships supplying capacity and technological capability infused into each portfolio company; and direct ownership, which removes the need to sell the conversion, negotiate scope, or win an internal champion, since it can simply be directed.

Why it matters

The claim reframes what a buyout firm's edge can be during a platform shift. The conventional private-equity lever set is financial engineering, multiple arbitrage, procurement, and management change. This is a claim that the lever is a proprietary conversion technology, and that the shift itself is where the return lives,¹ even as some investors, including Brad Gerstner, read the same enterprise-software correction as a reason to stay away. The two views are compatible: if you believe you own the conversion capability, a derated asset is an input rather than a warning.

Tensions

No return figure has been given for the second version specifically; thirty companies are producing revenue, but the amount, margin, and multiple are undisclosed, and the two-and-a-half to three times figure belongs to the first version and is being borrowed as a prior.¹ The first version's precedent may not transfer cleanly, since on-premise-to-cloud was a delivery-model change with a mostly mechanical cost structure and a well-understood target state, while generative AI conversion changes the product itself, and the target state differs by industry, workflow, and regulatory regime. A company converted per week is a fast stated pace, which either means the conversion is shallower than a full business transformation, or that the factory is genuinely industrialized; "producing revenue associated with this sort of conversion" is a low bar that a single shipped feature would clear. Direct ownership also solves the adoption problem and creates a selection problem, since a firm that can direct a conversion can also direct one that should not happen, with no customer able to refuse to buy and signal that. The claims are, throughout, self-reported by the seller to an audience of prospective investors.