Advice on AI worth repeating

Be a good manager (even if you don't manage anyone).

08 September 2026

I’m often struck by how much my day-to-day experience working with AI differs from that of the people I meet in jobs outside of tech – be they in other industries, in the public sector or in civil society.

And when people ask me for advice I often come back to the same things, so I’ve written them up here in case it’s useful for a wider audience.

All of what follows is about how you choose to interact with AI. Most tools have a natural language interface, and it turns out many of the same principles and techniques that you would use to guide a team member or collaborate with a peer can be very productive in this context too. It’s probably no coincidence that some of the most effective people I know when it comes to working with AI are also good partners and people managers.

NB: This post is primarily for people doing knowledge work: roles that typically involve a lot of problem solving, communication and self-direction. It assumes basic familiarity with using AI tools at work, for example summarising documents or getting answers to simple questions. It’s not written for more technical users, so if you’re interested in vibe-coding your own apps or getting OpenClaw running on a spare Mac Mini that’s a different conversation!

1. Try talking

Writing a great prompt can be daunting, particularly when you’re trying to articulate a complex request or aren’t quite sure where to start. Switching to voice input is a great way to reduce friction. You can dictate naturally and lean on the AI to parse what you’ve said for the key points, or grab an app like Superwhisper that will tidy up your voice input first before sending it on for the AI to work with.

2. Ask to be interviewed

Even the most capable AI tools can’t read your mind. A good prompt goes a long way, but still it can be hard to capture an exhaustive description of what you want done. Instead try giving the AI a rough idea of what you want, and then get it to interview you: have it ask you 10 or so questions it thinks will help it clarify your request / requirements and then use the answers to inform its approach.

3. Make a plan

Today’s AI tools are pretty good at executing complex, multi-step tasks. They can also be quite keen to get going, which is great lots of the time but can also result in misdirected effort. I will often work with the AI to define the problem, and then explicitly ask it to make a plan and present this to me for review before diving into the detail. It’s much easier to course-correct here than later!

4. Give good feedback

AI increasingly does a good job but I still refine the final product to suit my taste – sometimes working with the AI on edits, other times taking the raw output and making changes manually. Either way I always close the loop: I give the AI back its original version and my final version, then I ask it to compare the two, think about the differences and remember lessons to apply to future work.

5. Prioritise yourself

Last but by no means least, AI can help with your own growth and development. Busy jobs tend to squeeze out time for reflection on what you’ve achieved, how you’ve shown up and what you’ve learned. So at the end of every Friday afternoon my AI initiates an interactive, guided discussion that gets me to focus on being a better leader. When we’re done we write insights to a learning log that I can consult later, and make a note of any actions or commitments to check in on next time.