AI in Software Development: Are You Building the Right Thing?
If you’re driving in the wrong direction, the worst thing you can do is get a faster car and turn the headlights off.
I’m watching a lot of teams trying to drive really fast in the dark right now.
Software teams can now build in an afternoon with the lights off what used to take a whole sprint. What they build works and does exactly what it says on the tin. The problem is nobody checked what the tin says.
Every Thursday I play five-a-side football. It’s the first 10 to put their name down from a WhatsApp group of 30 people. Each week someone had to chase who’s playing, who’s out, who’s paid and what the teams are.
In July, while teaching myself to build with AI, I made a simple app. Players mark themselves in, reserve or out. Captains pick teams live on their phones, and paste the teams back into WhatsApp.
Within 20 minutes of me sharing the link, my brother Affy had broken it by starting the draft before we had 10 players. I fixed it and sent a new link. Graeme started calling him our chaos monkey.
A week later, after the teams had been posted, players began copying the message back into the chat with “Paid” typed next to their names. Alim asked for payment tracking in the app and I added it the next day.
Raz asked if the link could stay the same each week so we could pin it in the group. That was in the app seven hours later.
Then, disaster happened, the venue cancelled our regular booking due to low attendance over summer. To win it back, we agreed to pay for seven games upfront. I built the app for a one week interval and it no longer worked.
Graeme posted: “One for the product backlog - AS A player I WANT to be able to log if i can attend for multiple future weeks SO THAT i can add dates when i already know i can’t play.”
I told him it was a perfect user story and asked for acceptance criteria. He replied: “Ac1- make it work :)”
The seven-week version went live the next morning. The 10 regulars are booked in by default. If one of them drops out, a reserve claims the spot, and the app pastes the update “Aman replaced Joe” for the group.
Not all requests made it. When Raz asked if notifications were on the backlog, I said I’d need budget approval and a GDPR compliance manager.
The group uses the app every week now, and I couldn’t have designed it in advance, no single person could have. The paid column came from a workaround in the group chat. The seven-week block came from a booking problem and the single URL was born out of pure laziness. It ended up being the right app because people used it and told me what was missing.
My question is what we do with the time AI gives us back. I’d spend it getting smaller pieces of value in front of real users and checking if it’s any good.
It’s not about two-week sprints anymore, iterations should be based on feedback latency. The feedback loop on my app was measured in hours. How often does your team hear from someone who actually uses what you build?
Are you building faster, or learning faster?
If AI is accelerating your delivery, it’s time to rethink how your teams get feedback, test ideas and deliver value that matters.
Let’s talk about closing the gap between building software and knowing you’re building the right thing.
Comments