Long Lake, an applied AI holding company, wants to buy established service businesses and run AI inside them instead of selling AI to them. Prathik Naidu laid out the argument in a long post on X, published on 21 October 2026, writing in the first person plural for the firm. It is a pitch for an investment model from the people running it, and it deserves to be read that way.
His starting question is what stays costly once machine intelligence gets cheap. The answer, he says, is work where customers trust a provider with the whole outcome, where human judgment is required, and where regulation decides who is accountable. His examples are fixing an urgent leak rather than writing a note about it, owning a complicated tax decision rather than pulling data from documents, and handling a whole trip, delays included, rather than booking the flight.
Naidu then walks through why an outside vendor struggles. A working agent still has to clear security review and connect to systems that can be twenty years old. Someone has to get staff to use it, and someone has to settle whether it is priced by seats or by tokens. His claim is that models are not the bottleneck. Organizations are, because real decisions run through veterans, exceptions, and unwritten know-how that no ticketing system records.
The money argument is where the post gets specific. Suppose a vendor lands a $1 million annual contract and its software lifts the client’s operating earnings (EBITDA) by $5 million annually. At a 10x multiple, Naidu says, that is $50 million of enterprise value, and nearly all of it lands on the client’s books. Those figures are his illustration, not results from a deployment. Building a rival firm from scratch fails differently: no customers, no reputation, and no easy way to hire specialists.
Ownership, he argues, fixes five things: access to the experts, live data on whether an agent got a task right, aligned incentives, inherited trust, and cash flow to keep investing. He calls these a compounding advantage and compares the model to holding companies such as Berkshire Hathaway, Danaher, and Constellation.
On scale, the post says Long Lake has raised $7.5 billion, with D1, Elad Gil, Alpha Wave, and General Catalyst among the backers. It says 40 businesses are now partners, spread over home services, architecture, HR, and further fields, and that the team numbers 60. It also cites its Amex GBT deal, valued at $6.3 billion, as the entry into corporate travel, and says it looks at over 5,000 acquisition targets a year. All of that is Long Lake describing itself.
What the post leaves out is the evidence. It describes a J-curve, with spending first and gains later, but gives no revenue per employee, margin change, or payback period from any acquired firm. It also does not say what multiples Long Lake pays. That matters because a buyer takes on a price that already reflects the business as it was, so the AI gains have to beat the premium.
The argument is still a useful signal about where capital is heading. Naidu’s own logic implies a threat to AI vendors selling into services firms: the next call an owner takes may come from a bidder instead of a salesperson, and a firm weighing an AI contract now has a sale as one of its options.
Prathik Naidu, in a post on X, 21 October 2026.
