AI agents for small business: a plain-English UK guide

What is an AI agent, in simple terms?

Three things get called “AI” at work, and the difference between them is who decides the next step:

• A chatbot answers. You ask, it replies, and nothing else happens until you ask again.
• A workflow follows fixed steps: when an email arrives, copy these details into that sheet. Anthropic, which makes Claude, describes workflows as models and tools “orchestrated through predefined code paths” (Building effective agents).
• An agent chooses its own steps. In Anthropic’s words, agents “dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks”.

For most small businesses, a workflow is the right first step. An agent earns its place when the steps change from one case to the next.

Agentic AI vs generative AI

• Generative AI makes things: text, images or code, in answer to a prompt. ChatGPT, Claude and Gemini are examples.
• Agentic AI uses that ability to act: it can look things up, use other software and decide what to do next, within limits you set.
• Most agents you can use today are a generative model with tools, instructions and limits added. The model writes; the agent does.

How many UK businesses use AI in 2026?

The Office for National Statistics (ONS) asks UK firms about AI in its business survey. The latest figures:

• 30% of all businesses reported using at least one AI technology, and 38% of businesses with ten or more employees, in the survey that ran from 21 September to 4 October 2026 (ONS, October 2026).
• Among firms with ten or more employees, use rose from around 12% in late 2023 to around 35% in June 2026 (ONS, July 2026).
• 29% of businesses with fewer than ten employees reported using AI technologies in September 2026 (ONS).
• Text generation using large language models, the kind behind ChatGPT and Claude, was the most used AI technology, at 21% of firms with ten or more employees in September 2026 (ONS).
• Of the firms with ten or more employees that use AI, only 10% said they use it extensively, in June 2026 (ONS).

The ONS calls these results official statistics in development, so treat them as a guide to the trend rather than an exact count.

Five small-business jobs an AI agent can take on

• Sorting the shared inbox: labelling each email, drafting a reply, and leaving it for a person to check and send.
• Turning emails and PDFs into spreadsheet rows, so nobody re-types them. Our guide to turning PDF documents into Excel shows one free way.
• Keeping a tracker of jobs, orders or enquiries up to date from the emails and forms that come in.
• Chasing missing documents with polite, dated reminders, and stopping as soon as they arrive.
• Writing a weekly summary of what came in, what was done and what is still waiting.

Where a person stays in charge

An agent should never have the last word on anything that is hard to undo. Keep a person approving:

• Anything sent to customers, suppliers or the public, until you trust the drafts.
• Money: payments, refunds, and anything that commits the business to a figure.
• Decisions about people, such as who is hired, offered work or turned down. The ICO’s report on automation in recruitment, Recruitment rewired, says that where employers rely on a person being involved, it must be “applied consistently to all candidates within a hiring stage”.
• Access: give the agent only what its task needs, such as one mailbox or one folder, and read-only access wherever that is enough.

How to try an AI agent safely this week

1. Pick one task that happens every week and follows roughly the same pattern.
2. Time it by hand for a week, so you have a figure to compare against.
3. Write its rules on one page: what it should do, what it must never do, and when it should stop and ask. Your first automation, on one page shows how.
4. Run it with approval switched on, so a person checks every action before it happens.
5. After two weeks, compare the time it takes with your figure from step two, and check a sample of its work.

AI agents for recruitment agencies

Agencies have plenty of repeat admin that follows clear rules, which suits agents. Three places to look first:

• Client portals: the updates your team re-types into a client’s system. See stop re-typing into client portals.
• Compliance documents: reading each upload and holding the ones that do not match, so your team opens only those. See only open the documents that need you.
• Hiring decisions stay with a person. Our one-page policy on where AI helps in hiring shows where to draw the line.