Spec is the New Code

What is a team worth without trust? (Rhetorical question - don’t answer that please.)

I’ve been thinking about this for 12 years as a Project Manager and Agile practitioner. But I never expected AI to be the thing that taught me the lesson all over again.

I’ll admit it - I was a boomer when it came to AI. When it became the “it” word, I assumed it would be like the Cyber era - where everything got the word cyber added to it (relevant or not). But the noise got loud enough that I had to see for myself.

I gave the AI an idea to work on together, made it clear that the last time I’d hands-on developed anything, Gangnam Style was topping the charts, and asked it not to tell anyone how old I am.

At first things looked easy - setup instructions, guiding questions, tool suggestions. But as we progressed the red flags appeared. It stopped keeping to decisions we’d already made, bugs kept repeating, and the whole thing started to diverge from our main goal. When I challenged it, it always apologized and suggested a fix - but honestly, it felt more like a fight than working together. I couldn’t trust anything it did.

There’s your answer to my opening question.

I lost my forward momentum until one day I had to help my daughter with her homework. As the person she (allegedly) got her short attention span from, I thought there has to be a better way - so, like Wile E. Coyote, I went back to the drawing board. But this time with my professional hat on.

Before a single line of code, we discussed how we were going to work: the methodology, where to log decisions, and how to avoid losing context. We defined the different agents we’d need, the specific skills they required, the technology we’d use, and the risks we might face. Was it overkill for a kids’ game? Probably. But I kind of liked it.

Then came the Spec - and this is where I truly saw the power of AI. We spent a long time refining goals and generating ideas. The “How” we’d agreed on clicked into place as the agents challenged me when we diverged from our goal, suggesting alternatives that worked better. It felt like a real team.

We followed planned stories (I let the AI decide what they were, I’m not a monster). I got a report after each cycle. My Jira board stays updated so I always know my status even if work stops mid-session.

It didn’t feel like a battle anymore.

I know a sample size of one isn’t enough to make an informed decision - but my hands-on experience leads me to think that while Agile emphasizes small user-stories over full upfront specs, shifting left and focusing on the Specification and Design phase provides real value and speed when moving into AI.

How are you building trust with your AI teammates? Specifically, do you trust them enough for a full Spec, or do you still find yourself needing those fast, continuous feedback cycles? I’d love to hear your thoughts.