The FCA Just Cleared the Runway. It Didn't Take the Controls.

Nikhil Rathi spoke at Agents of Change on 24 June, and it would be easy to file the speech under "Rethinking regulation for the age of AI." The more useful reading is what it tells you about how the FCA now sees its own role. Three things came through clearly. The regulator is becoming more proactive, more collaborative, and more openly pro-competition. Underneath all three sits a message for firms that is sharper than the optimistic tone suggests. Let me take the posture first, then the part that matters most for anyone deploying agents to make decisions.

The FCA is openly rethinking what an effective regulator looks like when technology moves faster than any rulebook can. Rathi was blunt that legislation will never keep up. The response is not to wait for it. He pointed to Buy Now Pay Later, which took around six years to come formally inside the regulatory perimeter, while the FCA did not wait six years to intervene. Expect more of that pattern: action ahead of legislation where consumer harm is in view.

The regulator is also turning the technology on itself, exploring agentic AI as a "first responder" to monitor wholesale markets, sifting roughly a billion rows of data a day to catch market abuse faster. A growing part of the job, in Rathi's framing, is stewardship as much as supervision. Helping firms and markets navigate change, and acting early rather than waiting for certainty. That is a regulator choosing to move first.

The second signal is how plainly the FCA wants to build with the industry rather than at it. The Supercharged Sandbox lets firms test on real data and compute alongside partners like Nvidia, and soon Google. The AI Lab now carries a dedicated Agentic Academy. The AI Consortium brings firms and the Bank of England around the same table. Rathi's own line was that the FCA's approach "can't be designed first, consulted on later," and that the hard problems, data availability and integration among them, are best understood by the people running the systems.

That is a regulator asking to co-design, not dictate. The same instinct shows up in its cross-sector work on fraud with Ofcom under the Online Safety Act, after UK losses of nearly 1.3 billion pounds to payment fraud last year, much of it originating on social media and messaging platforms. Collaboration is becoming the operating model, not the exception.

Third, competition runs through the whole speech. AI lowers barriers to entry, and the FCA explicitly welcomes that. Its role, Rathi said, is not to protect incumbents but to make sure competition works in the interests of consumers and the economy. Some firms will rise fast, others will fall behind, and the regulator signalled it is comfortable with a transition that feels bumpy as long as market integrity holds. Expect its system-wide and competition powers to be used more routinely, as part of the toolkit rather than as exceptional interventions.

There was a great deal I am skipping here. Tokenisation took up a large part of the speech, including approval of the UK's first natively tokenised authorised fund, and there was substantial material on operational resilience and critical third parties. All of it matters, and all of it is for another issue. The thread worth pulling today is what this confident, open posture quietly asks of the firms it is encouraging to scale.

When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.

89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.

Start flowing free

Because for all the openness, the FCA did not relax its grip on one thing. Accountability. Rathi was precise that as agentic systems move from supporting decisions to coordinating and transacting, accountability for regulated activities and outcomes must remain clear. That is not a new requirement, and that is exactly the point. The FCA's innovation team has been making the same argument publicly in recent weeks, that responsibility for AI-assisted decisions already sits inside the regimes firms operate under today, including senior manager accountability. Nobody is waiting for a new AI statute to decide where the blame lands. It is assigned already.

Read alongside the pro-innovation tone, the message becomes clearer. The regulator is clearing the runway. It is not taking the controls. Scaling is encouraged. Owning the consequences of what you scale is assumed.

There is a second tell, and it is easy to miss. Rathi asked that these systems be "designed with the right human oversight." That phrase tells you the FCA is not all-in on fully autonomous agents, at least not yet. It wants the speed of autonomy with a human anchor, partly because, as the speech acknowledged, investors and consumers will be wary of delegating real decisions to systems they do not understand. Oversight is the price of trust, and trust is the thing that lets adoption scale at all.

Here is where responsibility lands squarely on firms, and it is worth being honest about how much. The regulator is not going to publish a universal rule for how much autonomy is too much, because, as Rathi himself said, a rule that specific would be out of date before the ink dried. So the balance becomes a firm-level judgment. How far do you let the tool run on its own, and how much human judgment stays in the path before a decision takes effect. That dial, between aggressive deployment and meaningful oversight, is yours to set and yours to defend if a regulator, a court or a customer ever asks.

And setting it once is not enough. The lazy version of human oversight treats it as a switch. A human is named, a box is ticked, the control is declared to exist. Oversight is not a switch. It is closer to a battery, and it drains. Practitioners deploying these systems inside large banks have started measuring how fast, and the picture is uncomfortable: reviewers who scrutinise carefully in the first weeks and are clicking through by week twelve, with the rate at which they overrule the machine quietly falling, not because the agent improved but because the humans stopped looking closely. A firm that cannot tell those two situations apart does not really have oversight. It has the appearance of one. The FCA asked for the real thing.

Every missed call is a job your competitor just booked. Podium's AI Employee responds in under 2 minutes, qualifies the lead, and schedules the job — while your crew keeps working.

See It In Action

None of this is an argument to slow down, and the FCA is clearly betting it is not needed. The opportunity is real, the regulator is leaning in, and the firms that scale well will be the ones that treat accountability and oversight not as friction to be minimised but as features of a deployment worth trusting.

That is the space this newsletter exists to watch. As AI adoption accelerates across regulated financial services, The Adoption Gap will track how the sector scales it, stays resilient, and earns the confidence of the consumers it ultimately serves. The runway is open. Stay with us as we watch who takes off, and how well they fly.

This newsletter tracks the distance between what AI promises in financial services and what firms can actually deliver. If a colleague forwarded this, you can subscribe now.

Find me on X and LinkedIn if you want to continue the conversation there.

For sponsorship enquiries, email [email protected]

Regulatory commentary is for informational purposes only and does not constitute legal, compliance, or financial advice. Consult a qualified professional for advice specific to your firm. BuildMoat product mentions are editorially assessed and disclosed where a conflict exists.