Nobody prompts anymore. They build loops.

Something changed this year in how AI tools are built, and it is worth understanding even if you never build one yourself.

People used to write prompts. You write an instruction, hand it over, and get one answer back. The newer tools do not work that way, and the people building them will tell you they have mostly stopped writing prompts at all. They build a loop instead. They give the AI a goal and some tools, and it runs on its own: it acts, looks at what happened, decides the next move, acts again, and keeps going until the job is done or it gets stuck.

In a practice, prompting is like asking a junior to pull one figure and report back. A loop is handing them the whole job, reconcile this client's quarter, and letting them work through it without checking in between each step. One gives you an answer to review, and the other does the work and acts as it goes.

That is why "AI agents" are suddenly everywhere, and it leaves every firm with one question I want to spend the next few weeks on. Where are you happy to let the loop close on its own, and where does a human still need to be the one who decides?

You already know how to think about this, because you do it with people. You would let someone you trust draft a letter unwatched, but you would not let them file a return or move a client's money without a second pair of eyes.

That is the lens for the series. Not whether AI is good or bad, but where the loop gets to act, and who stays responsible when it does.

This is Post 1 of Series 2 (The AI That Acts) in The AI Footnote, my weekly note on AI in UK accountancy.

Lexendo

Find the risk across your client portfolio, before HMRC does.

Six domains of UK tax and compliance coverage. 12,580 cited UK sources. Audit Intelligence mapped to ISA 240, 315, 520, 550 and 570. 30-day money-back guarantee.

If Lexendo doesn't find a risk worth acting on in your first 30 days, full refund.