Initial reactions: Building & working with an agentic PM

Everyone asks if AI can write code faster. Almost no one asks about the other job on an engineering team – keeping the board honest, chasing follow-ups, linking tickets, tracking the dependencies everyone forgets.

That's the PM's work, and it's what quietly decides whether a team is aligned or just guessing.

So we built an agentic project manager and put it on a real dev team. Ahmed Bashir and Shivam Gupta break down the first few sprints – honestly. What we built, what it took over, what surprised us, and where it still needs a human.

The idea is simple: a developer just does the work they love. The agent keeps the record up to date behind them and picks up the tasks that usually get dropped.

And dropped work is the expensive kind. It doesn't announce itself like late work does – it just quietly makes everyone plan off a picture that isn't real. Weeks later, the bill shows up. Catching it early is the agent's whole job.

What you'll take away:

– How developers just build, while the agent keeps status updates from being a tax on their time.
– What it actually took to bring an agentic PM onto the team and into the sprint rhythm.
– How it catches the follow-ups, stale tickets, and unlinked work before they turn into wrong understanding.
– Where it earned trust, where it didn't, and where humans still have to lead.