Most of what’s written about using LLMs (ChatGPT, Claude etc) focuses on teaching it to think better. We spend time refining prompts, shaping context, and improving how a model reasons.

Lately, I’ve been experimenting with the opposite: using AI to help *me* think better. To engineer context for myself.

In fractional leadership, you move between very different environments, multiple days each week. Each company has its own rhythm, language, and mental model. Getting back into that headspace quickly, without the ramp up time of a typical 5-day work week, is tricky.

So I’ve been trying something new. At the end of each client day, I’ll record a short voice note, just a few minutes capturing what’s on my mind, what decisions were made, what’s next, and what I might need to remember next time.

I drop that into ChatGPT, ask it to structure and summarise it, and paste the result into next week’s (private) calendar invite.

When that day comes around, I read the summary, skim Slack and email, and I’m usually back in flow within half an hour. It’s early, but it’s working surprisingly well.

Back in my full-time days, I’d write a to-do list for the following week on a Friday. This helped a bit, but the combo of voice note + AI summary feels faster, more comprehensive, and weirdly, more human. Like I’m having an async conversation with my future self. It captures the tasks, but also the thinking, mindset and concerns that are often missing from the clarity of a tight task list.

I could imagine this being useful beyond fractional work too. Some examples that spring to mind:

→ Before going on holiday, when you want to clear your mind but help your future self rebuild momentum when you return.

→ Before an all-day conference or workshop, when you’re deep in a project and want to pick things up again easily afterwards.

→ Even before a long weekend, when you want to properly switch off without losing the thread.

Any situation where you need to offload and reload context might benefit from this kind of experiment.

I’m sure there are smart ways to use agents or automations to asynchronously gather and report progress well too, but I haven’t got there yet. Anyone else experimenting with AI in this way and have ideas or tactics to share?