You have probably tried a chatbot
You typed a question, got a decent answer, and then closed the tab. Maybe you asked it to draft an email and ended up copying and pasting the text yourself. Useful, in a limited sort of way.
Now picture the same idea with hands: it sends the email for you, updates the spreadsheet, chases the invoice, and remembers what it did yesterday. That is the difference between a chatbot and an AI agent.
A chatbot answers questions; an agent does work
A chatbot sits behind a text box and waits. You ask, it replies. The conversation ends when you close the window. It has no memory of your business, no access to your tools, and no ability to act on anything you discussed.
An AI agent is built to operate inside your actual workflow. It can read your inbox, write to your CRM, pull data from a spreadsheet, and take scheduled actions without you being there. The shift is from a tool that talks to a tool that works.
What an agentic harness actually does
The language model on its own (the thing generating the text) is a brain in a jar. It cannot send an email, check a calendar, or remember anything from one conversation to the next. An agentic harness is the scaffolding that turns that raw capability into something useful for a real business. Here is what the harness adds.
Memory means the agent remembers your business across conversations. It knows your standard reply templates, your pricing, which clients are on which projects, and what you decided last Tuesday. That context means you are not starting from scratch every time you interact with it.
The harness connects the agent to your actual software: email, calendar, CRM, spreadsheets, your website. The agent can read and write to these systems directly, moving information between them without you copying and pasting anything.
You can teach an agent repeatable procedures. For example, when a new enquiry arrives, check if the postcode is in your service area, reply with the standard quote template, add the job to the CRM, and flag anything unusual for the owner. Once taught, the agent follows the same steps every time.
The agent does not need you to be online. It can run at 6am, checking overnight enquiries, preparing a morning summary, and sending it to your phone before you have poured your first coffee. Scheduled work is what turns a helpful tool into something that actually reduces your workload.
The agent also asks before it does anything irreversible. It will not send an email without approval, commit to a price without confirmation, or message a client without you reviewing the draft. The harness enforces these rules so the model cannot run ahead of you.
A simple running example makes this concrete. Imagine a small building company. An enquiry lands on their website at 11pm.
The agent picks it up, checks the postcode against the service area, replies with a quote template and a link to the booking calendar, adds the contact to the CRM, and flags the lead for the owner to review in the morning. None of that required anyone to be awake. A chatbot could have drafted the reply text. It could not have done any of the rest.
Why this matters for small businesses
A lot of UK firms have experimented with AI. The UK SME AI Adoption Report 2026 found that 54% of UK businesses use AI in some form, but only 11% use it deeply. Most of what they use is chatbots and content generators. They are trying AI without AI doing any real work for them.
An agent that acts on your behalf closes that gap. It is the difference between having an intern who can tell you what to do and having an intern who actually does it.
Small businesses can adopt this faster than large ones because there is less organisational friction. You do not need a compliance committee to teach an agent how to handle your enquiry workflow. You just need someone who knows how to set it up.
Self-hosted and open source: your data stays yours
Most AI tools on the market run on someone else’s servers. Your business data sits inside a vendor’s cloud, and you are renting access to a black box. When the pricing changes or the service shuts down, your workflows go with it.
An open-source agent harness works differently. The software is free to use and can be read and audited by anyone. It runs on your own server, so your data never has to leave the country. There is no vendor lock-in and no subscription rent on the core software.
Hermes Agent is one example of an open-source agent harness. It is built by Nous Research, free to self-host, and works with Slack, email, and other channels.
SumFactors deploys and manages exactly this kind of system for UK businesses on UK servers, tailored to each business’s tools and processes. This is not a one-size-fits-all platform you sign up for. It is a framework you adapt to how you already work.
See what an agent could take off your plate
We deploy and manage AI agents for UK businesses, built on open-source Hermes Agent and hosted on UK servers. If you would like to see what an agent could handle for your business, with no cost and no commitment, we will walk you through it. Read more about our AI agent for business service. If you want to understand what AI marketing actually costs for a UK small business, we have a straightforward guide for that too.
