For two years the question about AI in investment management was "what can it do?". The question I'm hearing in the room now is "who signed it off, and can you prove it?"
The questions in the room have changed
A couple of years ago, most of my conversations with founders selling AI into hedge funds and asset managers were about getting the meeting at all. Then it was about the demo. Then, as I wrote in July, it was about adoption and getting clients to actually use what they'd bought.
The conversation I'm hearing more of now starts somewhere else. Not "what can it do?" but "who's in control of it, who signed it off, and what happens when it gets something wrong?"
October gave that question some real weight.
Using more, trusting less
The Bank of England and the FCA published their latest survey of AI in UK financial services at the end of October (Bank of England). 86% of the firms that responded now use AI, up from 75% in 2024. 80% have AI agents deployed. Only 2% have given those agents broad autonomy.
Two days later, at a Bloomberg event in New York, some of the biggest names in hedge funds said much the same thing (Hedgeweek). Citadel talked about keeping its agents inside strict controls. A Bridgewater partner called keeping autonomous agents from creating security or operational risk "an unresolved challenge". Man Group's CIO warned against "cognitive surrender", people relying on AI output so much they stop thinking for themselves.
So the firms using AI the most are also the ones keeping it on the shortest lead. I don't read that as scepticism. It's what grown-up adoption looks like in a business where one bad output can move real money.
The regulators are writing it down
In the same month, both sides of the Atlantic started putting rules around it. The SEC's exam priorities for next year swap "artificial intelligence" for "super intelligence". Examiners will look at how firms monitor its use in trading, AML and the back office, and whether the AI claims in their marketing stand up (InvestmentNews).
The FCA says it will build a framework for agentic finance with the industry and the AI labs, covering agent identity, delegated authority and accountability. Firms will be expected to report significant AI failures and near misses (FinTech Global).
None of it is finished. But if you're a COO or a CCO at an asset manager, you now have a good idea what you'll be asked. And you'll want your vendors to help you answer it.
The vendor ends up holding the pen
This is the part I'd point any founder or CRO selling AI into this market towards. The same Bank of England and FCA survey found 53% of the AI systems firms rate as high-materiality are now bought in from third parties, up from 35% in 2024. For systems built in-house, firms rely on knowing how they work. For bought-in systems, 81% rely on vendor assurance and contracts instead, and 54% say applying model risk management to vendor models is a challenge.
Put simply, the buyer is trusting the vendor to show them it's safe. That changes what gets a deal over the line: audit trails, permissions, a human sign-off step, and a straight answer on what the model can and can't touch. A couple of vendors in our market have already had their AI management systems certified against ISO 42001 for exactly this reason.
Governance used to be a section near the back of the security questionnaire. More and more, it's part of the product.
Which changes who you need
It changes the people as well. A salesperson who can light up a portfolio manager with a great demo is still worth a lot. But the deal now has to get past model risk, compliance and procurement too, and they ask very different questions.
Across the market, the people getting harder to find are the ones who can sit across from a CCO and hold their own. Solutions engineers who understand how a fund is governed, not just how it trades. Product people who think about audit and permissions before anyone asks. Sellers who know a deal can be won in the demo and lost in the risk committee.
That isn't a new type of person. It's what the best buy-side sellers have always done. There just aren't many of them, and plenty of AI vendors are looking for the same people at the same time.
Closing thought
The first wave of AI selling was about showing what was possible. The next one is about showing it's under control. The firms using AI the most are the ones keeping the tightest grip on it, and the vendors who make that easy are the ones who'll still be in the room when the risk committee meets.
When did a buyer last ask you to prove it, rather than show it?
Photo: LUFANG CAO on Unsplash