← All posts

AI agents are everywhere in 2026. Here's what one actually does for a small business

"Agent" is the word the whole industry is shouting this year. Strip the hype and it's simpler than it sounds — and more useful in some places than others. Here's the plain-English version from someone who builds them and works a sales floor.

Open any tech feed right now and you'll drown in the word "agent". New agent frameworks, agent harnesses, agents that write code, agents that book your holidays. The engineering world is having a genuine moment with it. The trouble is that none of that noise tells a business owner the one thing they actually want to know: does this do anything for me on a normal Tuesday?

This is the honest version of that answer. What an agent really is, the handful of jobs where it earns its keep on a small team, and the places where it's the wrong tool no matter how impressive the demo looked.

What an agent actually is

Forget the sci-fi. A normal AI tool answers one question and stops — you ask, it replies, done. An agent is the same intelligence, but given a goal and the ability to take a few steps on its own to reach it: check something, decide what to do next, do it, then check again.

Here's the difference in one line. A chatbot tells you how to write a chase message. An agent notices a customer has gone quiet after approval, drafts the right kind of chase, and flags it for you to send — without being asked. Same engine. The agent just has a job and a short leash, instead of only a mouth.

That "short leash" bit matters more than the intelligence. A useful agent for a small business isn't one that runs your company while you sleep. It's one that handles a narrow, repetitive job end to end and hands it back to a human at the point where judgement is needed.

Where an agent earns its place

Three jobs where I've actually seen this pay off, rather than just demo well.

1. The thing that should have happened, but nobody chased. Every business leaks money through follow-ups that didn't get made. A quote that never got a nudge. An approval that went cold. An agent watching your pipeline can spot the gap and prepare the next action — the message, the reminder, the task — so the human just approves it. On a car-finance floor, that's the difference between a deal that closes and a customer who drifts to whoever called them back first.

2. Turning a messy input into a clean output. A voicemail, a forwarded email, a screenshot of a customer's details. A person normally reads it, understands it, and types it somewhere tidy. That "read it and sort it" step is exactly what an agent is good at: pull the details out, structure them, drop them in the right place, and stop for a human to check. Boring, repetitive, and gone.

3. Answering the same internal question for the hundredth time. "Will this lender take a zero-hours contract?" "What's the message we send when someone ghosts after approval?" On most teams the answer lives in one experienced person's head, and everyone queues at their desk. An agent that knows your rules can answer that instantly, for everyone, without the bottleneck. The senior person gets their afternoon back.

Where an agent is the wrong tool

The hype skips this part, so I won't.

Anything with real consequences and no human check. Sending money, promising a customer a figure, making a decision you'd have to unpick later. An agent can prepare these. It should not be the one to pull the trigger. The moment a mistake is costly and irreversible, you want a person's name on it.

A job you can't describe. If you can't explain the steps to a new starter, you can't hand them to an agent either. Agents don't bring order to a vague process — they carry out the vague process faster, mistakes and all. Sort the process first.

A task that happens twice a year. Same rule as any automation. The payback comes from repetition. If it's not something your team does most weeks, the effort of building the agent will outrun the time it saves.

So what should a small business actually do

Not launch an "AI agent strategy". That's the version that ends up as a slide nobody actions. Do the small, real thing instead.

Pick one job your team does every week that's all reading, sorting or chasing and no real judgement. The most annoying one. Give that single job to an agent, keep a human on the approve button, and live with it for a month. If it quietly removes the annoyance, you've found where this technology actually belongs in your business — and you've learned it for the price of one small build, not a transformation programme.

That's how the tools I build have always grown. Not from a grand plan about agents, but from one repetitive job on one sales floor, handed to a system, kept on a short leash, and extended only when real people asked for more. The word on the feeds will change again next year. The test won't: does it remove real, repeated work, and does a human still hold the wheel where it counts?