By Elton Boocock, founder of Thinkivity, providing AI consultancy and training exclusively for the UK glazing industry.

Last month, the column was about what AI sees in your sales data — patterns, gaps, the things you stop noticing when you’re inside a business every day. The honest takeaway was that AI is good at showing you what your numbers say, even when you’re too busy to look.

This month, something that’s been on my mind. AI is moving past the stage of just showing you things. It’s starting to actually do them.

For the last two years, the conversation has mostly been about AI as a writer. Drafting an email, polishing a quote or creating content in marketing, even summarising a meeting. All useful. All within a chat window where the AI hands you something and you decide what to do with it.

That’s quietly changing. ChatGPT now has Agent Mode. Anthropic has Cowork. Microsoft is rolling out agentic features inside Copilot. The shape of these new tools is different. You don’t ask them to write something. You ask them to do something. Open a website. Pull together a list. Update a spreadsheet. Sit on the screen for an hour and complete a task in the background.

I’ve been spending time with these tools, watching where they work and where they don’t. The honest answer is they’re not finished. They get stuck. They make odd choices. They work better on some tasks than others. But what’s striking is how quickly the gap is closing between what an AI can describe and what it can actually carry out.

Here’s where it starts to matter for our sector.

A glazing business has a long tail of small, repetitive jobs that nobody really enjoys but somebody has to do. Tools Like Business Pilot help a lot for installers, but what about outside of that?

Updating a customer record after a phone call. Adding a summary of a Teams conversation to notes, working out where to place marketing spend. None of these tasks need experienced judgement. They need attention, accuracy, and time – which is exactly what we don’t have enough of.

AI agents that can sit at a screen and do those tasks change the maths. Not in a dramatic way. In a quiet way. The job that takes someone 40 minutes on a Tuesday morning gets done in the background while they’re talking to a customer. The follow-up list that gets neglected because there’s never quite a moment for it gets put together overnight, ready for review first thing.

Let’s be clear, these tools are not yet autopilots. They make mistakes, and the mistakes look confident. You wouldn’t hand them a task that involves a customer’s money or a contractual decision. You shouldn’t hand them anything where a wrong answer creates a bigger problem than no answer. If anything, this is where AI will make things worse, not better.

The right starting point is tasks that are repetitive, low-risk, and clearly bounded. Update this list. Check this report. Draft this digest. The kind of work where the worst case is, “you have to redo it,” rather than, “you’ve upset a customer.”

The other thing worth saying is that the businesses I see getting value from this aren’t the ones that try it for an hour and give up. They’re the ones who pick one task, work through the rough edges, and then expand from there. The first time you ask an AI agent to do something for you, it will probably struggle. The third time it will be quicker than you. The tenth time you’ll wonder why you ever did it manually.

For business owners and managers, the question worth sitting with isn’t, “what can AI do for me?” It’s narrower than that. What is the one job in my week that I would happily hand off to a careful, slightly slow assistant who never gets bored, never gets distracted, and is happy to be checked? Start there. The tools are now good enough that the question has a useful answer.

This isn’t an end point. The agents will get better. The tasks they can take on will get bigger. The line between “tool” and “team member,” which we touched on a few months back, is going to keep blurring. What’s different now is that the conversation has moved out of the lab. You can use these tools this week, on a real job, with real data, and see for yourself.

Less play. More purpose. That was the line a few columns ago, and it still stands. AI is at its most valuable in our sector when it stops being a curiosity and starts being a quiet contributor. Writing was a useful place for it to start. Doing is where it’s heading next.

If you’ve been waiting for AI to feel less like a chatbot and more like a colleague, the next 12 months will be the most interesting we’ve had so far.