A scaling drinks brand we've worked with told me they were holding off on AI until they'd nailed down a proper strategy. Sensible enough, on the face of it. A few weeks later they were back in touch, because everything had moved on and the strategy was stale before anyone had signed it off.

That's the pattern I keep seeing. What AI can do changes month to month, sometimes week to week. Spend three months building a strategy and you've built something that's out of date by the time it's finished.

It isn't a discipline problem. These are people who've written a decent five year plan before and had it work in other contexts, the tool just moves faster than the document.

So don't build a strategy. Build a set of operating principles.

Not a policy pack. How you as a team decide to work with AI, which holds up even as the tools underneath keep changing. From the businesses I've worked with, three of them matter most.

The obvious pushback is that principles are just a strategy with fewer pages which is fair to an extent. The difference is what happens when the tools change. A strategy has to be rewritten. Principles don't, because they're about how you decide, not what you buy.

Before anyone on your team touches a tool, get in a room and agree a handful of things. You don't need a policy document, just an agreement on five questions.

This should be a conversation, not a deliverable. Give the questions in advance and nobody needs to go away and research it. Everyone in that room already has a view on most of these questions - gather opinions and write down the answers to these principles.

Freeing up your team isn't just permission to try things. It's time. Real time to test and to learn, whether that's a proper course or an afternoon messing about with a tool. Not something people are squeezing in at 9pm after work.

Then pair it with two things: a way to check whether what they're trying is actually working, and a way to tell each other about it.

One brand I work with does this really well. Each person picks one thing to focus on, tests a few options, and once they find what works they spend two to three weeks properly embedding it before moving on. They're not running five experiments at once and losing track of all of them.

The sharing matters just as much. Another brand I work with has set up a Slack channel just for AI wins, so everyone can see what's working for each other. It's a low effort thing that does a lot of work, because the alternative is three people solving the same problem separately and nobody finding out.

That Slack channel isn't just for the team. It's a live feed for leadership on what's actually working, which is how you spot the thing worth rolling out to everyone rather than leaving it with the one person who found it.

Once you've found it, make it the standard way of working, not a nice-to-have one person does. Give someone clear ownership of keeping it maintained and checking the ROI still stacks up six months on. Tools change, and what worked in March might need a tweak by September.

The brands that get stuck aren't the ones that moved too fast. They're the ones that spent the time writing the plan and then had nothing left over for the part that's genuinely hard, which is getting people to change how they work.

Set your limits, free up your team, scale what works. Then keep cycling through it, because the tools underneath will keep moving whether you've written anything down or not.

Put an hour in the diary with whoever needs to be in the room, and answer the five questions. Write the answers in one shared doc where anyone can find them and revisit them regularly

Sanjay is co-founder of Pesto, AI-native ops software for UK food and drink.