Prove It
A founder spent a workshop this week bragging that 95% of his company's code was written by AI. The follow-up question was about test coverage. Nobody answered.
Kirby Montgomery tells that story as a diagnostic, not a joke. He and Jason Gill run technical diligence at Code & Co., and the silence told them what they needed: no automated testing, and a technical leader saying what he thinks buyers want to hear. The claim founders are proudest of has become the thing buyers most need proven, in the same year the claims got free. Asked which diligence document he no longer trusts, Gill did not pick one. "Everything we get is a summary made by Claude. It all got the same cornflower blue. It all looks the exact same." His fix is to stop reading summaries and ask for the raw data underneath.
What survives that filter is the opposite of what a data room leads with. "All the stuff that you thought wasn't important is now even the most important because people can't replicate that pain," Montgomery says. Gill's example is a company that did not consider itself an AI business until diligence found more than 25,000 human written support tickets in its own records, built over a decade of integration work. Scar tissue is the defensible part now, because everything cleaner can be regenerated. For anything a seller does call proprietary, the standard is blunt: "our smoking gun is if someone says proprietary, prove it."
The evidence cuts both ways. Reviewing one engineering team's own productivity tracking, the pair found AI-written code taking ten days longer to approve than human-written code, at a business whose sponsor model assumes engineering headcount stays flat. Caroline Carman, a senior associate at Rallyday Partners, put the gap in an associate's terms: AI is "really good at getting to like the top 10 things you need to know about this industry really quickly," but "I don't think the AI is persuasive yet. just informational." Retrieval is solved. The argument is not.
The fund side spent the week on the same burden in a different room. Colleen Fay, who leads financial services at Withum, reduced the fees and expenses standard to five words: "disclose what you do and do what you disclose." Per the panel reading Private Funds CFO's survey of 132 US managers, 90% said their most recent exam included questions on fees and expenses, while only 8% of exams produced a deficiency on those topics. That is the profile of a subject that has moved from differentiator to hygiene. Bill Myers of Private Funds CFO on where it ends: "even as the 2 parties argue over the ceilings that they want to put on regulation, the floor keeps rising." Fay's read on the investors is sharper. They "may not always agree on the outcome, but they expect to understand the rationale behind it."
Which makes the week's retail number worth sitting with. The same survey found 73% of managers have no plans to accept retail or semi-retail capital. Jay Ritter, who has tracked IPO data since the 1980s, was asked about opening 401(k) plans to private equity and landed in the same place from the other side: "I don't see a free lunch sitting there." The managers worth buying are not the ones asking for the money. "Sequoia capital is not looking for retail investor money. They're turning away institutions, even though they charge premium fees."
Sources: Interviews from Private Equity Funcast (Aug 26, Jason Gill and Kirby Montgomery of Code & Co.), Private Equity Funcast (Aug 24, Caroline Carman of Rallyday Partners), Private Equity Spotlight (Aug 24, Colleen Fay of Withum and Bill Myers of Private Funds CFO), and Run the Numbers (Aug 27, Jay Ritter of the University of Florida).
Technical diligence now treats AI-written code as a claim to be tested rather than a selling point, and the same burden of evidence is reshaping fee and expense disclosure to LPs.