JPMorgan and Vista Equity: the agentic AI business is not in the models, it's in the software that distributes them

By J.P. Morgan Private Bank · September 6, 2026.

JPMorgan Private Bank has devoted its "Alternatives Access" webcast series to explaining to its private banking clients why software is, in its view, the soundest way to invest in the agentic artificial intelligence wave. The conversation brings together an executive from the bank itself with Ashley McNeill, head of equity capital markets at Vista Equity Partners —one of the largest software-focused private equity managers— and co-chair of Vista One.

The figure that frames the entire argument: according to JPMorgan, nearly 90% of companies with $100 million or more in revenue are privately held and do not trade on public markets. In software the percentage is even higher. McNeill puts the share of software companies that remain private at more than 95%, against the small sample of listed brands that the public mistakes for "the whole market." The reason, she explains, is that software goes through phases of rapid growth —changes in product, in commercial strategy, in control over customer data— that sit poorly with the mandatory quarterly reporting of a listed company.

The second block of the conversation is about the market: after the speculative peak of 2020-2021, public software valuations have returned, according to the analysis McNeill presents, to their ten-year historical average. In other words, the correction already happened before the current AI spending cycle, which the firm uses as an entry argument: the sector does not depend on an AI bubble to justify its price, but on already known fundamentals —more than 80% of public software revenue is recurring or contractual, according to the data cited in the webcast.

But the most relevant point for understanding where institutional money is looking is the phased thesis McNeill lays out on the deployment of AI: a first phase of building infrastructure and semiconductors, a second of hyperscalers that enable the technology, and a third —where she places the investment opportunity— in which software acts as the "adopter," the application layer that actually brings AI to companies and consumers. It is an idea we have been documenting at Zendoric from another angle: the fight over AI is waged less and less over who has the smartest model and more and more over who controls the distribution, the integration and the workflow data that connect that model to a real customer.

McNeill sums it up with an image worth reading closely: software is splitting into a "two-tier" economy. The first tier is the agent that directly replaces a human task —she gives the example of a restaurant reservation handled by an AI assistant that, she says, "works 24/7"—; the second is the underlying software that still provides the base solution on which that agent runs. Put bluntly: the first layer is the one that absorbs the labor cost savings; the second is the one that capitalizes those savings as recurring revenue for its owners.

That distinction confirms, in the words of a firm that invests billions in software, what we at this publication have been pointing out sector by sector: administrative and routine service tasks are the first to disappear, not because a model is "smarter" in the abstract, but because someone packages that capability into a product that replaces a payroll with a subscription. The cost savings that agentic AI promises are not shared only between companies and consumers: a significant part is captured as margin by the software that orchestrates the agent, and by those who invest in that software before it reaches the public markets.

It is worth keeping in mind what this material is: promotional content from a private bank, aimed at high-net-worth clients, to justify interest in evergreen and semi-liquid investment vehicles in software private equity —structures that, unlike ten-year closed-end funds, allow investors to enter and diversify across years without being locked into a single investment vintage. The bank itself warns at the outset: these are the speakers' opinions, not a guarantee of future results. The "good entry point" thesis serves, among other things, to sell that access.

That said, the underlying fact matters beyond the financial product that wraps it: when private capital specialized in software structures its AI bet this way —infrastructure first, hyperscalers next, enterprise application last as the place where monetization really happens—, it is describing the same map of power we had been anticipating. In the short term, that means more automation of administrative and customer service work, with the real risk that much of the productivity gain stays in the software provider's margin before reaching the worker or the consumer. In the long term, if that application layer truly manages to bring down the cost of services that are expensive today because of their labor component —from administrative management to healthcare requiring little clinical judgment—, it is exactly the kind of abundance that underpins our thesis: not a larger language model, but thousands of concrete applications absorbing routine tasks to free up human time for what does require judgment, relationships or presence.