There’s a phrase you’re going to keep hearing in 2026: “AI agents.” It’s already everywhere. Tech companies are racing to build them, consultancies are selling them, and most business owners are quietly wondering what they actually are and whether any of it is relevant to a company with 20 employees in Northampton.

The short answer is yes. But not in the way most of the hype suggests.

The Next Step After Automation

If you’ve already automated parts of your business, even something as simple as an automatic invoice reminder or a form that feeds into a spreadsheet, you understand the basic idea: set up a rule, and the system follows it. Every time. No thinking required.

An AI agent is the next step. Automation follows rules. Agents make judgement calls.

That’s the simplest way to think about it. An automated workflow says “when X happens, do Y.” An AI agent says “when X happens, assess the situation, decide what the best response is, and do it.” The difference is that the agent can handle variability. It can read context, weigh up options, and pick the right action from a set of possibilities.

This isn’t science fiction. It’s not five years away. Businesses are building and using these systems right now, and the cost is a fraction of what most people assume.

How an AI Agent Actually Works

An AI agent has three core components, and none of them are complicated to understand.

Context. The agent is loaded with information about your business. Your processes, your policies, your products, your customer history, your tone of voice. This is what makes it useful rather than generic. A general AI tool gives you generic answers. An agent with your business context gives you answers that sound like they came from someone who actually works there.

Judgement. Using that context, the agent can assess incoming information and make decisions. Not random decisions. Decisions based on the rules and patterns you’ve defined, combined with the flexibility to handle situations that don’t fit a rigid template.

Tools. The agent can take action. It can send emails, update records, create documents, move data between systems, notify people. It doesn’t just think. It does.

The crucial part, and this is where most of the fear around AI falls apart, is that you decide how much autonomy the agent has. For some tasks, it runs independently. For others, it prepares everything and waits for a human to press the button.

The Email Agent: A Real Example

Here’s a concrete example that makes this tangible.

Imagine every email that arrives in your business inbox gets read by an AI agent. Not a spam filter. An intelligent system that understands your business.

The agent reads the email, categorises it (new enquiry, existing customer, supplier, complaint, invoice, general question), and routes it to a specialist sub-agent. Each sub-agent has the specific business context it needs. The customer enquiry agent knows your services, your pricing, your availability. The supplier agent knows your current orders and delivery schedules. The complaint agent knows your resolution policy.

That sub-agent drafts a response. Not a template. A genuine, contextual reply that reads like it was written by someone who knows the situation. Then it drops that draft into the relevant person’s draft folder.

A human reviews it, tweaks it if needed, and hits send.

The AI did 90% of the work. The human kept 100% of the control.

When people see this for the first time, three things tend to surprise them equally. First, that this is possible today, not in some future product roadmap. Second, that a human still reviews everything before it goes out. And third, that it doesn’t cost enterprise money to build. This isn’t a system reserved for companies with 500 employees and an IT department. A business with 15 people can have this running.

The Business Assistant: Your Team’s New Colleague

The email agent is powerful, but there’s another use case that changes how a business operates day to day.

Picture an AI assistant that every member of your team can talk to. Not a generic chatbot. An assistant that’s been loaded with your business context: your policies, your procedures, your FAQs, your product details, your pricing, your customer guidelines.

A new starter asks “what’s our returns policy for trade customers?” and gets an accurate, detailed answer immediately. No waiting for a manager. No digging through a shared drive. No asking the person who’s been there longest.

A project manager asks “what’s the status of the Henderson order?” and the agent checks your systems and gives a current answer.

A team member asks “can you draft a response to this supplier about the delayed delivery?” and the agent writes one using your tone of voice and the details from the order.

But it goes further than answering questions. You give the agent tools. It can update your CRM, create calendar entries, send notifications, generate documents. It becomes a team member that never sleeps, never forgets a process, and never needs to be told the same thing twice.

For a business with 15 to 30 employees, this changes the dynamics of institutional knowledge. The information that usually lives in one person’s head becomes accessible to everyone, instantly.

Who Controls Your Business Data?

Here’s where this connects to something worth thinking carefully about: where your data actually goes.

Most AI tools on the market are fully cloud-based. Your customer data, your processes, your business logic all live on someone else’s platform. If you stop paying, it’s gone. If they change their terms, you adapt or leave.

With a self-hosted automation setup, the picture is different, but it’s worth being precise about what that means. The automation infrastructure, the workflows, the databases, the business logic, and the stored data all live on your servers. You own and control them completely.

The AI component is different. When an agent needs to make a judgement call, it sends a request to an AI provider like Anthropic or OpenAI via their API. That means some data does pass through their systems during processing. However, there are two important distinctions. First, you control exactly what gets sent to the AI. You decide which data the agent includes in its requests. Second, API providers typically don’t store or train on your data, unlike consumer AI tools where everything you type may be used to improve their models.

The practical difference matters. With a fully cloud-hosted platform, all your business data lives permanently on someone else’s servers. With a self-hosted setup, your data stays on your infrastructure and only specific pieces are sent to the AI when needed for processing, with nothing retained afterwards.

As the ICO’s recent guidance on agentic AI makes clear, businesses using AI systems that make decisions affecting individuals need to understand where data flows and who controls it. A self-hosted approach gives you a much clearer answer to that question than a fully cloud-based alternative.

Is Your Business Actually Ready?

This is the honest part. AI agents are powerful, but they’re not magic. And most businesses aren’t ready to deploy them on day one. Not because the technology is too complex, but because the foundations need to be in place first.

Your processes need to be clear. If a human can’t explain how a process works step by step, an AI agent can’t follow it either. The agent is only as good as the instructions and context you give it. Businesses that have documented their key processes, even roughly, are in a much stronger position than those running on tribal knowledge.

Your data needs to be accessible. If your customer information is scattered across six different platforms with no connection between them, an AI agent can’t pull it together. The agent needs to access your data to make useful decisions. That means your systems need to be connected, or at least connectable.

Your team needs to understand what’s happening. The biggest barrier to AI adoption isn’t technical. Research from DSIT found that 46% of UK businesses cite lack of knowledge as their primary barrier, and 32% simply don’t understand the benefits. That’s nearly half of UK businesses held back not by budget or technology, but by not having someone to show them what’s possible and how it applies to their specific situation. “AI will replace my job” is still the first thing many people think, even though the reality is closer to “AI will handle the tasks you wish you didn’t have to do.”

This is exactly why we exist. We don’t just build automation. We help businesses understand what AI can do for them, show them how it works in practice, and build it on infrastructure they own. If you’re in that 46%, you don’t need a bigger IT budget. Get in touch so we can show you what AI actually means for a business like yours.

The practical path is straightforward. Get your processes in order. Connect your systems. Start with one agent doing one thing well. Then expand.

Three Things That Surprise Everyone

People who haven’t seen a working AI agent tend to carry three assumptions that turn out to be wrong.

“This must be years away.” It isn’t. The tools to build business-specific AI agents exist today. Claude, the AI model we use, can be given business context, connected to your systems, and deployed as a working agent. Not a prototype. A production system handling real work.

“AI means no human oversight.” The opposite. Well-designed AI agents are built with human checkpoints. The email agent drafts a response and waits for approval. The business assistant suggests an action and asks for confirmation. You set the level of autonomy. Full automation for low-risk tasks, human review for anything sensitive.

“This is only for big companies.” The economics actually favour smaller businesses. A company with 500 employees has an IT department that can build custom tools. A company with 20 employees has the founder doing admin at 9pm. The ROI of an AI agent is proportionally higher for the business where the owner’s time is the most constrained resource.

Where to Start

If this resonates with how your business operates, here’s the practical path forward.

Pick one pain point. Don’t try to deploy AI agents across everything at once. Choose the task that eats the most time, causes the most friction, or relies too heavily on one person’s knowledge. Email triage, customer enquiry handling, and internal knowledge access are common starting points.

Document the process. Before any technology gets involved, write down how the process works today. What comes in, what decisions get made, what goes out. This becomes the foundation the agent works from.

Connect your systems. If the process involves data from multiple tools, those tools need to talk to each other. Automation handles this, connecting your existing software so data flows between them without manual copying.

Keep your data on your infrastructure. Build the automation, workflows, and business logic on systems you own. The AI processing itself uses providers like Anthropic or OpenAI via API, but you control what data gets sent and your business information stays on your servers between requests. That’s a very different picture from handing everything to a cloud platform permanently.

Start with human review on everything. Let the agent handle the preparation. Let your team handle the approval. As confidence builds, you can increase autonomy for routine tasks. But start cautious.

If you want to see what an AI agent could look like for your business, drop me an email at [email protected]. No jargon, no pressure. Just a clear conversation about where an agent would save you the most time and what it would take to build one.

The businesses that move first on this will have a significant advantage. Not because the technology is exclusive, but because they’ll have months of refined business context and optimised processes while their competitors are still wondering whether AI is relevant to them.