Building my own AI agent made me a better skills coach

Until recently I was a digital skills coach at a Manchester university, supporting degree apprentices in digital technology solutions and user experience. Every three months I met each apprentice and their line manager to talk through wellbeing, where they were up to with the requirements of their programme and progress against their knowledge, skills and behaviours (KSBs), then I wrote it all up as a formal review. This was meaningful work, but took 40 to 45 minutes to get done.

Over the past year I’ve built a review write-up agent in Copilot. It analyses the meeting transcript and produces a structured draft, aligned to KSBs and quality expectations, which I then refine into the final document.

My write-up time is now down to 18 to 20 minutes, more than 26 hours saved every three months across my caseload. That’s welcome, but my real excitement has come from the effect it’s had on my working practice.

Knowing the session would be transcribed changed how I run the meetings. I’ve become more deliberate in how I structure them, and I take fewer notes, knowing the agent will read the transcript. Because I’m not half-focused on writing, I’m more present in the room, which means better developmental coaching and pastoral support.

As one apprentice I work with put it: “I’ve noticed in our more recent reviews that our conversations certainly have flowed more freely in a greatly productive and natural manner without the need to pause or capture written points.”

Building the agent has sharpened my sense of how to capture impact and quality in my write-ups too. I don’t see it as doing my job for me, I see it as helping me get better at what actually matters: my ability to support and coach apprentices.

This matters beyond my own practice, too. Ofsted’s inspection guidance for FE and skills providers, in place from September this year, is explicit that its focus is the quality of decision-making about AI, not simply whether providers use it. How thoughtfully we use these tools is now part of how quality gets judged.

I didn’t develop that instinct for what good quality looks like on my own. Around a year ago I joined an AI working group at my university. We experimented with basic Copilot and talked about our worries and our hopes about the ethics and practicalities of using it in reviews.

Our head of quality endorsed and facilitated the sessions, which gave us the confidence to speak openly. Talking to colleagues there shaped how I thought about efficiency and about capturing impact in our roles.

When I later got access to Copilot’s premium tier, I was encouraged to build my own agent, which pushed me to look harder at how I conducted reviews myself, and at what counted as good evidence of quality, not just what the agent produced. The group had a steer on quality from our central team, but real autonomy within that. Its purpose was to establish whether there’s a business case for the university buying licenses for all skills coaches, grounding us in what the role needs, not just what the technology can do.

FE Week opinion writers have argued that AI adoption succeeds when it’s employer-led, from the top down. My experience points to something that sits alongside that: if agents like mine are adopted more widely, they’ll likely be standardised through Copilot Studio. But the working group stage that got me there matters in its own right, building a culture where skills coaches feel able to explore what AI can do for their practice, rather than being handed a finished tool.

Skills coaches are best placed to know what’s worth exploring and how to do it responsibly. But that only works with endorsement from senior leaders, so exploration gets recognised, not just tolerated. Concretely: give practitioners protected time and a sanctioned space, like a working group, to design their own workflows, not just training on someone else’s finished agent. That’s where the gains that don’t fit in a spreadsheet come from.