AI Can Write the Roadmap. It Can’t Run the Room.
Substack Live recap — Aadil Maan × Colette Molteni
AI didn’t kill the technical program manager. It just took away the busywork that was hiding the actual job.
This was the thread running through my conversation with Aadil Maan, who writes The Art of Doing Technical Program Management and has spent almost sixteen years watching organizations decide, over and over, whether a new tool is a threat or a mirror.
Aadil’s take was blunt: half of what gets pointed at AI doesn’t need AI. It needs someone willing to ask what problem they’re actually solving before reaching for a solution. He’s watched leadership’s reflex shift from let’s fix the process to can AI solve this — and that reflex is costing organizations the one step that would have made the fix stick.
Not because AI is weak. Because the question was never about AI in the first place.
What most people miss is that a technical program manager’s real output was never the Jira ticket or the status report. It’s synthesis — every stakeholder’s input, compressed within a narrative someone with less time and less context can act on. Aadil put it plainly: TPMs are the glue. The lightning rod for engineering’s speed, leadership’s impatience, and product’s shifting asks, all landing on one person who’s expected to hold the shape of the whole thing.
So when the inputs themselves start changing, when fifty demos replace one PRD, when code that took three sprints now takes an hour, the glue work doesn’t disappear. It gets harder to see. And it gets more important, not less, because someone still has to judge what’s signal and what’s just a well-formatted hallucination.
This is where empathy stops being a soft skill and becomes infrastructure.
Aadil’s diagnostic for a program going sideways wasn’t technical at all. Every program, he said, is either a people problem or a technology problem — sometimes both. Psychological safety. Trust between functions. Whether someone felt safe enough to raise their hand before the problem became unfixable, AI can accelerate the technology side of that equation. It cannot manufacture trust between two teams who’ve stopped believing each other.
Speed isn’t always progress. Sometimes it’s just friction happening faster.
The generalists win in this next stretch — not the specialists who bet everything on one technical lane, but the people who’ve absorbed enough pattern recognition across companies and contexts to know when something reads well and means nothing. That discernment, Aadil argued, is the new toolkit. Not access to the model. Everyone has that now. The differentiator is knowing what to do with what it hands you.
Especially in fast-moving environments like tech, that judgment doesn’t come from a prompt. It comes from having sat in enough rooms to know what dysfunction sounds like before it shows up in a status report.
AI can write the roadmap. It can summarize the sprint, draft the PRD, generate the fifty demos. What it can’t do is stand in the room when engineering is exhausted, product is anxious, and finance is asking questions nobody’s ready to answer — and hold that space without losing the thread or the people in it.
That’s still the job. It always was.
