The Warhorse Is Already in the Building

There is a phenomenon in medicine called the Herxheimer reaction. When the right antibiotic finally kills the bacteria, the body floods with toxins from the dying organisms and the patient gets acutely worse before they get better. The cure presents, briefly, as the disease. Doctors who don’t know what they’re looking at sometimes stop the treatment.

It feels like this is what AI is doing to SaaS companies right now. SaaS multiples that peaked at circa 20x revenue in 2021 sit at single digit multiples. The public software cohort has shed hundreds of billions of dollars in market cap. Blue Owl is down over 40% this year on its software exposure. The phrase on Wall Street is SaaSpocalypse.

It isn’t terminal, it’s a Herxheimer, the discomfort is the cure landing.

The show pony Like everyone else, the public market has watched frontier models do impressive things in thirty seconds and concluded, reasonably, that a lot of horizontal SaaS is now a thin UI sitting on a problem an LLM can solve directly. It is real. It does some of what enterprise software used to do, faster and cheaper, but there is the argument that it is a bit of a show pony. Security, support, a systematic approach, user security, integrations, consistent calculations, they all need to be managed.

What the market hasn’t priced The market has marked portfolios down because it has seen the show pony but it hasn't seen the warhorse standing next to it.

PE software-heavy portfolios are sitting on the exact assets that know exactly what to do with AI technology. People who build systems for a living. Proprietary data nobody else has. Customer relationships measured in years. Domain expertise embedded in the codebase of capabilities that is an organisation. Technology stacks that are imperfect but real, and crucially, already integrated into the workflows of the customer and cannot easily replace.

It might be slower to light up AI across an enterprise codebase than hand rolling some data and building a front end dashboard, but if you think AI is powerful in the hands of a finance or legal person, wait until you see what the software people can do with it. The show pony quickly turns into a warhorse.

This isn’t a hopeful argument, it’s a structural one. Frontier models multiply the productivity of people who already know how to ship software. The constraint on AI value isn’t access to the models; everyone has that. The constraint is technologists with discipline, working inside organisations with real data and real workflows. PE owns more of those than any other capital pool in the market and doesn’t yet appear to have noticed.

The warhorse isn’t AI. It’s the combination of frontier AI and people who already know how to systematise things at scale.

And all of this brings me back to private equity.

The cohort best placed to lead There are two opportunities here, and they sit on top of each other.

The first is internal. PE firms are full of smart people working on hard problems with serious money on the line. The job is analytical, the data is rich, the decisions matter, and the feedback loops are real. If ever there was an industry that should be showing the rest of the world how AI ought to be deployed at scale in an institutional setting, it’s PE.

But being well placed isn’t the same as being well organised. PE getting good at AI doesn’t mean a hundred well-meaning partners prompt-engineering their way through diligence on the weekend. It means treating AI the way the firm already treats capital; allocated deliberately, deployed against a thesis, measured against returns, governed properly. Cleverness scattered across an organisation is just noise. Discipline applied across an organisation is an edge.

What does this actually look like? Relentless structured data underneath every portfolio company. Relentless monitoring against the deal thesis, not the calendar. Relentless early warnings on covenant drift, customer concentration, working capital, key-person risk. Relentless analysis of how many portfolio companies have adequate pipeline coverage with the right revenue mix to hit plan, surfaced this morning rather than reconstructed at the next QBR. Relentless reuse of the same secure, audited substrate across diligence, monitoring, value creation, and exit. Relentless, because the whole point is that the work doesn’t stop and start. It just runs.

That isn’t a dashboard. It’s an operating posture. And it’s only available to a firm that has done the unglamorous work of treating its data, its security, and its AI infrastructure as serious capital expenditure rather than a series of one-off prompts.

And this isn’t an either/or. The clever prompts don’t stop. They get compounded. Amplified. Supercharged. The same prompt running over structured data instead of a hand-rolled export, inside a SOC 2 environment instead of a personal ChatGPT tab, citing sources instead of inventing them. The associate who impressed the room on Monday is still impressing the room ; just doing it on every portfolio company at once, on Tuesday, in a way the CFO can audit.

The second is external. A firm that has actually built the foundations can speak about AI to its portfolio companies with authority. Not as theory. As practice. PE has spent decades exporting financial discipline to the businesses it owns. The next decade is the chance to export AI discipline the same way.

The funds that get this don’t just operate better. They become the operating model. Their portfolio companies inherit it. Their LPs notice it. The market eventually prices it.

The funds that figure this out will quietly buy the warhorses in their own portfolios at pony prices, while the rest of the market is still sorting one from the other.