Turn AI From a Risky Expense Into a Product Advantage

Mike and I are settling in to run a webinar with Barclays Eagle Labs on how to integrate AI into your product without lighting cash on fire. The 'AI Question' is one that comes up in nearly every conversation we have with founders right now, and the honest answer is that most teams are getting it wrong in one of three predictable ways:

Mistake One: bolting AI onto the wrong feature.

Mistake Two: paying for an enterprise model when a small one would do.

Mistake Three: freezing, waiting for the perfect moment that never comes.

Each of these mistakes is expensive, in its own special way.

This post is the short version of what we'll be walking through. Sign up for the webinar if you want the long version with Q&A, or grab the free playbook if you'd rather just read it.

AI is a competitive advantage. It's also a strategic risk.

That's the line we keep coming back to. At the moment we've got ...hang on, just checking the CRM... 16 active clients / product developments on the go at SmplCo, and every single one of those products has some AI inside it.

And here's the thing: the amount of AI embedded in each is not what matters. The most succesful are the ones that picked the right place to put it, in the right way, at the right stage.

A pre-seed founder treating AI like a Series-B scale problem will burn months on infrastructure they don't need yet. A scale-up bolting on a chat interface because their competitor did will end up with a feature nobody asked for. The cost isn't always money — sometimes it's six months of focus. It can also be security (if you're looking into chatbots, for example, have a quick search for 'prompt injection' and feel the chills.)

To be clear: we think AI is great. We use it all the time and, we're proud to say, to great effect. We've been picked by Figma as a global exemplar of AI-assisted product development, and we just won Lovable's international SheBuilds hackathon. AI is involved in almost every one of our projects.

But we've seen plenty of teams use it badly. There's been plenty of times we've coughed politely and suggested maybe the right answer was less AI, not more.

Don't start with AI. Start with a plan and a structure.

This lesson is crucial and sits at the core of the playbook. And it's the bit most teams skip.

People start prompting before they really know what they're building, AI makes something that looks like an app, and then the problems show up. Things like unclear journeys, messy data, design that drifts from screen to screen, and code nobody wants to touch the next morning.

The workflow we've settled on after 125+ products goes like this:

Plan → Figma → Claude Design → Claude Code → GitHub → Deploy.

So, for the love of all that is holy, plan first.

The boring questions are where the value lives:

• who is it for?
• what problem are you solving?
• what's the first useful journey?
• what does "done" actually mean?

Using that, you create a one-page build brief. Then it's onto building the journey in Figma, including the three states everyone forgets: empty, loading, error.

Then Claude Design turns the system into rules so the AI builds inside it, not beside it. Then it's on to Claude Code and - THIS IS CRUCIAL - you ask it to plan before it codes, like a good junior developer instead of a caffeinated raccoon with access to your repo. GitHub, deploy, README, env vars in the right place. Then a sanity check on data, security, who can access what, and so on.

And whenever you're doing anything, always have this mantra repeating in your head: 'Build what makes me different, integrate what doesn't.'

And before you add any AI feature, ask the question that decides everything: "Is AI the product, or just a feature?" They're very different commitments.

Want the detail?

There's a free 13-page playbook that walks the whole thing through, with the prompts we actually use and the patterns to avoid. Download it here — no sales pitch, just the framework.

If you'd rather hear it live with the chance to ask Mike and me anything, register for tomorrow's webinar. It's free, it's online, it's an hour, and there's a recording for anyone who registers but can't make it live.

And if you're already building something AI-powered and want a second opinion before you ship, get in touch. It's the kind of thing we are really rather good at, and we have the badges to prove it.