My AI Personal Assistant: Leading Through Voice Notes on the Walk to Work

Fifteen minutes that shape my day

My walk to school takes about fifteen minutes each way. For the last two or three years, that walk has become one of the most productive parts of my working day.

I talk to AI as I walk. In the morning I'm thinking through what the day needs from me. In the evening I'm chewing over what happened, reflecting and decompressing before I get home. Sometimes I'm drafting a letter or newsletter. Sometimes I'm just asking questions and getting it to do some research for me.

As the technology has improved, so has the process. What started as the occasional voice note has become a routine that genuinely helps me structure my day and get things done.

A personal assistant, not a replacement

The best way I can describe how I use AI is that it's like having a personal assistant. It isn't doing the job for me. The leadership, the guidance and the steer all come from me. What it does is go away, do some of the research, form a draft, and hand it back so I can edit and change it as I see fit.

Mostly, I'm using it to help me word things quickly, effectively and clearly. If you give it enough context and tell it what you want to say, it's very good at that. It isn't making things up for me; it's saving me the time that careful wording always takes.

I still choose to write some newsletters from scratch, and there are other areas where I deliberately don't start with AI. I still need to exercise my brain. At this stage in my career that matters a little less than it did earlier on, but I think it's worth protecting.

A typical morning: the newsletter

The other morning I needed a newsletter ready to go out by the end of the day. As I walked, I brainstormed everything I wanted in it and gave a bit of context for each item. Then I asked the AI to email it to me.

By the time I reached my desk, the email was waiting in my inbox, ready to be edited and sent. A job that might have hung over me all day was already well underway before I'd taken my coat off. As you will know, that plan to do one job first thing can often be side lined for the entire day. A busy small school requires us all to wear multiple hats.

Keeping it safe and simple

I don't trust AI with access to my own mailbox, so I set up a separate Gmail account just for the AI and connected it to that. It sends from its own account to me, never from mine.

I also never ask it to draft anything using anyone's actual name. To be clear, most of what I'm doing isn't sensitive anyway: newsletters, curriculum thinking, researching current initiatives, product research. But these simple boundaries mean I don't have to think twice about it.

Pushing actions forward

One of the most useful things I've found is purchasing research. If there's something we want to buy for school, I ask the AI to go off and find suppliers. It often turns up suppliers I hadn't already identified. It can compare pricing and interrogate the terms and conditions on their websites. This morning I had it researching different shelving for our art store, a problem that I had asked multiple humans to help me solve for years. The AI just got on with it and provided me with some options and a short list. Perfect.

Often I don't need to do any further work on it myself. I can forward the email to a team member with much of the groundwork already done. That's the real value for me: it helps push forward the actions I want to see happening in school.

Context makes it powerful

These days I capture everything on the walk: reflections, meeting notes, to-do lists. I've given the AI access to my previous chats as a kind of memory, along with documents where we keep notes. The more you use it, the more powerful it becomes, because it has more context to base responses on. Whenever I'm drafting something, it brings forward the relevant information straight away.

Cutting through the bureaucracy

This is where it has made the biggest difference. Like many schools, we work within processes that can be burdensome and bureaucratic. Some would argue they aren't designed to slow things down, but in practice they do, so that only the most wanted things get through.

More recently I've been giving the AI forms for non-sensitive work, such as writing a business case for funding. It can complete other people's forms as they intended, especially when they include guidance notes for each box. For a school of our size, without a large team to absorb this kind of work, that has increased our productivity massively.

A good example is a Data Protection Impact Assessment for a new piece of software. These are detailed. You need to research the company's practices and consider the data risks. A paid AI model can research all of that, fill in the form, and cite exactly where it found each piece of information. That makes fact checking quick and easy. A job that could take a couple of hours or more can be closed off in around ten minutes. It can also write staff a clear set of instructions on what can and can't be done with that software. Depending on your processes that form may need to go to someone else for further approval, but the work is well underway and will hopefully be completed more quickly as a result.

Policies are another area. I use AI to critique our policies, suggest improvements and proofread them. Nobody wants to publish a policy with inconsistencies or errors in it. I used to ask staff to proofread, but no one person catches every mistake I make. AI gives me another layer of checking that I really value.

When it goes wrong

It isn't perfect. I once asked it to build a presentation from information I'd given it. It very helpfully created graphics to support the slides, supporting the learning by dual coding. But I didn't proofread it carefully enough, and one slide didn't quite say what I wanted. I'd been through it, but I was rushing alongside a lot of other things and missed the nuance.

The lesson is simple: take a moment. Even with all your context, it won't get everything right. Ultimately, the responsibility for the quality of what I produce rests with me.

What I keep it out of

If something is sensitive, I don't upload it. Once you understand how these tools work, you realise there's a degree of risk. I've turned off the option for the provider to learn from my conversations (which I think you need the paid plans for), but I'm still using mainstream models.

For more private work, there are better options. Microsoft and other providers offer versions of recent models that run on discrete servers and don't train on your organisation's data. This is also what you will want to look for from providers that are using AI with any of your more sensitive data - I will do a separate article on this later. Those are a better option anything more confidential, although not entirely without risk.

That doesn't mean safeguarding is entirely off limits for the mainstream models. I have used AI for anonymised, no-names discussions, for example to talk through interpretations of Keeping Children Safe in Education. It's useful for thinking, as long as no identifying details go in.

If you want to start tomorrow

My main piece of advice is to set up an easy way to get the output of your conversation out of the chat. In the early days I built a custom GPT that would send me emails. That option has gone, but it's now much easier.

My current approach is to set up a separate email account for the AI, connect it, and then simply brainstorm as I work, walk or travel. When you're done, say "send that to my work email" and it will be waiting for you when you arrive.

Keep the list of addresses it can send to very short. I only give it my own, because I don't want it emailing anyone on my behalf. When I draft an email for someone else, I send it to myself first and forward it once I'm in school. That way I know nothing has gone out that isn't quite right.

Fifteen minutes doesn't sound like much. But used well, it means I arrive at school with my thinking already organised and my first jobs already underway.