Imagine spending an entire day writing a carefully researched and well-argued document. Then someone takes a few seconds to generate a hundred objections, criticisms and counterarguments with AI.

Of course, you can use AI to respond. And the other person can use AI to attack your response. Before you know it, you're both producing enormous amounts of material that neither of you has the time to read properly.

I've seen this happen in competitive tendering. AI makes it remarkably easy to produce more text, more analysis and more criticism. But I'm not convinced that producing more actually gets us anywhere. At some point, we're simply creating work for each other.

That experience changed how I think about using AI. I've started treating it as an incredibly fast junior colleague. One that can read hundreds of pages, investigate unfamiliar subjects, ask useful questions and challenge my assumptions. But also one that needs guidance, doesn't know the reality of my client's business and can be confidently wrong.

Based on my experiences with actual tenders, I've developed a practical methodology for working this way. It's not rocket science (pun intended). Nor is it a clever collection of prompts. It's a fairly straightforward process: understand the tender requirements, talk to the people who know the business, gather evidence, identify what's missing, work out a strategy and only then start writing. And once there's a draft, have AI try to tear it apart.

The important part is knowing when to let AI do the work, when to challenge it and when to make the decisions yourself.

In this talk, I'll take you through an AI arms race that got rather out of hand, explain what I learned from it and show how I've translated those lessons into a practical way of working.

I don't think the future of knowledge work has to be a dark factory, with machines producing endless amounts of output and hardly any people left in the building.

I'm much more interested in what people can achieve with AI that they couldn't achieve without it.