From Chatbots to Agents: The Biggest Shift in Business AI Since ChatGPT

Something Has Changed

Six months ago, I was copying and pasting code into ChatGPT, asking it to spot bugs and suggest improvements. It was useful — like having a second pair of eyes that never got tired. But I was still the one doing everything. I’d read the suggestion, decide if it made sense, switch back to my editor, make the change, test it, and move on.

Today, I’m working with AI agents that sit inside our development environment and write code directly alongside me. They read the files, understand the context, make changes, run tests, and flag problems, all without me switching between windows or copying a single line. The jump in capability happened fast. Genuinely fast. And it caught even us off guard.

I’ve worked in IT for over 30 years. I’ve lived through the shift from dial-up to broadband, from on-premise servers to cloud, from manual provisioning to automation. None of those transitions moved at the speed this one is moving.

So What Actually Are AI Agents?

The simplest way I can explain it: a chatbot answers questions. An agent does work.

Ask a chatbot to write you a report and it’ll produce a draft. You then need to copy it somewhere, format it, review it, send it out. The chatbot’s job ends the moment it finishes generating text.

An agent goes further. It pulls the data itself, builds the report, drops it into the right system, and distributes it to the people who need it. It can operate software, follow multi-step processes, and handle the kind of routine work that currently eats up your team’s day.

That’s the fundamental shift. AI has gone from being a tool you use to a colleague you manage.

We’re Already Using This: Here’s What It Looks Like

This isn’t theoretical for us. We’ve been building agent-driven systems into our own operations for months, and the results have been significant.

We use AI agents to manage our infrastructure. They interact directly with our servers, network switches, firewalls, and backup software. When we built our new OpenStack private cloud environment, AI agents were involved throughout, configuring services, troubleshooting issues, and validating the deployment. That would have taken considerably longer without them.

We’ve automated threat response on our networks. We have workflows that detect suspicious activity, investigate the threat, and take appropriate action, isolating a compromised device, blocking an IP, alerting the right engineer. This used to require someone watching a dashboard. Now the first line of defence runs itself, and humans step in for the decisions that actually need human judgement.

Our monitoring has changed completely. We watch over Windows servers, Linux systems, network switches, firewalls, a wide range of kit across multiple client environments. AI agents now perform what I’d describe as human-like reviews of those systems. They don’t just check if a service is running. They research known issues online, cross-reference what they find against the alerts they’re seeing, and surface the relevant problems to the right team member with context attached. It’s like having a junior engineer who reads every vendor advisory and never forgets anything.

We’re helping UK manufacturers unlock data they’ve been sitting on for years. One of the most rewarding projects we’re working on involves a manufacturing client in the North West who captures huge amounts of operational data but has historically had no way to do anything meaningful with it. Using machine learning models, we’re now consolidating that data and highlighting patterns that directly impact product quality, things that were invisible before. And because we’ve built natural language interfaces on top of it, their shop floor team, people who aren’t technical — can ask questions of their own data in plain English. That was genuinely impossible for them twelve months ago.

What I’ve Learned About Working With Agents

Here’s something nobody tells you about AI agents: you have to manage them.

I don’t mean that in a vague, hand-wavy sense. I mean it practically. I’ve spent months learning how these tools think, where they’re strong, where they fall over, and how to give instructions that get results. It’s remarkably similar to managing a new team member, someone who’s capable but unfamiliar with your environment, your standards, and your way of doing things.

Get it right, and the output is extraordinary. Leave them unsupervised, and they will go off-task. Sometimes spectacularly. I’ve seen an agent confidently make changes across a system that were technically correct but completely wrong for the context. If I hadn’t been reviewing its work, it could have caused real problems.

The instinct is to let AI run and trust the output. Resist that. These are powerful tools, but they need direction and oversight, the same as any employee working through a complex task. The businesses that get the most from this technology will be the ones that learn how to brief, supervise, and course-correct their AI agents effectively.

MCP: The Bit Most People Haven’t Heard Of Yet

If you want to understand where all this is heading, pay attention to something called the Model Context Protocol, or MCP.

In simple terms, MCP gives AI agents a standardised way to connect to your existing business software. Your CRM, your project management tools, your accounting package, your website, MCP lets an AI agent interact with all of them through a single, consistent interface.

We’re already using this. Our AI agents connect to HubSpot, ClickUp, QuickBooks, and our own website through MCP. The practical effect is that AI can work across our business systems as if they were one joined-up platform, rather than a collection of separate tools that don’t talk to each other.

That’s a big deal. Most businesses run on a patchwork of different software. Getting data from one system into another usually means manual effort, CSV exports, or expensive custom integrations. MCP is starting to make that friction disappear, and it’s being adopted across the industry, from Microsoft to Google to open-source platforms.

Why This Matters for Your Business

I’ll be honest, the single biggest barrier we see to adoption isn’t the technology. It’s awareness. Most business owners we speak to have no idea how far AI has come in the last twelve months. They’re still thinking about chatbots when the conversation has moved on to autonomous agents that can run entire workflows.

And I understand why. The pace of change is unlike anything I’ve experienced in three decades of working in technology. Even we find ourselves constantly adapting, reassessing what’s possible, and updating our own skills to keep up. If you’re running a manufacturing business in Cheshire or managing a commercial estate in the North West, keeping track of AI developments isn’t your job, it’s ours.

The UK government clearly sees the opportunity. The Made Smarter programme is actively funding manufacturers to adopt AI and digital technologies, and the AI Opportunities Action Plan published in early 2025 is pushing hard for broader business adoption. But most of the SMEs we talk to haven’t heard of either. The support is there, it’s the awareness that’s missing.

So here’s what I’d suggest:

Don’t panic, but do pay attention. You don’t need to adopt agent technology tomorrow. But you should understand that it exists and that it’s maturing quickly.

Think about where your team spends time on repetitive work. Data entry, report generation, system monitoring, routine customer queries, updating records across multiple platforms — these are the tasks that agents handle well right now. If your people are spending hours on work that follows predictable steps, there’s probably a way to hand some of that to an AI agent.

Look at the tools you already use. Microsoft 365 is where most UK businesses live, and Microsoft is investing billions in agent capabilities through Copilot. If you’re already paying for a Microsoft 365 Business Premium licence, you’ll start seeing agent features appear without buying anything new. The same goes for Google Workspace and most major CRM platforms. The first practical step for many businesses won’t be a new purchase, it’ll be switching on something they’re already paying for.

Talk to someone who’s actually doing it. Not a vendor selling AI, but a team that’s using it day-to-day in real operations. The gap between marketing material and practical reality is wide, and you want advice grounded in experience.

Where This Is Going

The global analyst predictions are eye-catching, Gartner reckons 40% of enterprise applications will include task-specific AI agents by the end of 2026. Closer to home, a recent UK government report found that only 15% of UK businesses have adopted at least one AI technology, but that figure is climbing fast and the government’s target is to make the UK a global leader in AI adoption.

Those are big ambitions, and I’d take the specifics with a pinch of salt. But the direction is clear. AI is moving from a tool that generates content to a system that executes tasks. For UK SMEs, where teams are lean and every person wears multiple hats, that shift changes what a small team can achieve more than it does for a large enterprise with hundreds of staff.

For businesses our size and our clients’ size, this isn’t about replacing people, and in the current UK labour market, most SMEs can’t afford to lose the staff they have. It’s about giving your existing team capabilities they simply didn’t have before. The manufacturer who couldn’t analyse their own data can now spot quality issues before they hit the production line. The IT team that was drowning in alerts now has an AI doing the initial triage. The operations manager who spent Monday mornings pulling reports from Sage, their CRM, and two spreadsheets can have them waiting in their inbox.

That’s not science fiction. We’re building this now.

If you want to understand what AI agents could do for your business, have a conversation with us. We’ll give you an honest assessment, no jargon, no sales pitch, just practical advice based on what we’re seeing work.