A friend shared this with me the other day: Services: The New Software, which has an interesting take on who will get to claim the value from AI. He splits it on "intelligence" vs "judgement" and argues services are where the judgement, and therefore the moat, is, and that firms will start to sell the outcome rather than the platform that lets you get yourself the outcome (not accounting software, but your accounts all done). A director I used to work with at Quantium used to call that "the product of the product".
It's an interesting question though who ends up being best-placed to sell the product of the product over the next few years.
Three groups have a credible claim : the VC-backed AI-natives (eg. Crosby, Harvey, WithCoverage highlighted in the article, but also all of their direct competitors), the existing professional services firms (not just Big 4 but also the mid-market and solo practitioners across each vertical), or big tech, including the labs themselves, who now belong alongside Google and Microsoft on the big tech side by any reasonable reading.
The VC-backed group is the weakest of those three, which is the opposite of what you'd take from the piece. They don't natively have the domain expertise to deliver the professional advice (accounting or legal or whatever) at scale, outside maybe a founder or a couple of senior hires. Whatever they build can be copied and undercut by big tech, who pay for frontier model tokens at cost rather than at API mark-up meaning they can sustainably undercut them forever. What's more, copying features is software engineering: the longest bar on Bek's AI adoption chart (you'll have to click through via the link above), and big tech's traditional strength.
The professional services firms have an underrated advantage. Every matter they deliver is a corrected, signed-off output: an endless data source for boosting (in the Friedman sense, not just the regular sense) the output quality of their AI agents. When harnesses and frontier models commoditise, the moat becomes prompts. Writing good ones is hard - even for superstar practitioners. Services firms generate that signal at volume; the labs don't. It seems like they are big enough to buy any one firm, but there are too many firms across too many verticals for them to acquire their way in. They can't buy one of every type of firm in the economy.
This makes the recent announcements about services partnerships with PE firms quite clever. They have understood the value of that signal and are willing to give away a significant cut to get it. But those deals only let them compete, not guarantee victory.
There's also the question I discussed in The Defragged Economy about senior practitioners going independent. The two read as competing predictions but aren't really - they can coexist. Individuals move faster than software companies productising entire workflows - it's easier to do it at small scale, and they don't have the incentives alignment problem ("what's in it for me, why should I give up my expertise to be codified in AI bots") that companies do. But firms have other advantages - they have a lot more of that signal data being produced at volume, and a deep historical database, so even though they would be slower to start, once they get going the rate at which they can improve the quality of their output will surpass solo operators.
No doubt some firms will catch this and build their own loop, and others will lose their senior people before they do.