How I Use AI to Run Cross-Functional Programs at Google

The TPM's AI Inflection Point

When I first started at Google, managing a cross-functional program meant dozens of spreadsheets, hundreds of emails, and an endless cycle of status meetings. Today, AI has fundamentally changed how I approach every single one of those tasks — and I want to share what's working.

The key insight isn't that AI does the work for you. It's that AI handles the cognitive overhead so you can focus on the parts of program management that actually require human judgment: building relationships, navigating ambiguity, and making tough trade-off decisions.

Three Areas Where AI Changed My Workflow

1. Status Synthesis

Instead of spending 2-3 hours compiling weekly status reports from multiple sources, I now use Gemini to synthesize updates from docs, emails, and bug trackers. What used to take half a morning now takes 15 minutes of review and editing. The time I save goes directly into 1:1s with engineers and stakeholders — the conversations that actually unblock teams.

2. Risk Pattern Recognition

I've started feeding historical project data into AI models to identify risk patterns I might miss. When three different signals align — a dependency slipping, test coverage dropping, and eng velocity declining — the pattern becomes visible before a project is actually in trouble. It's like having a second set of eyes that never gets fatigued.

3. Meeting Prep and Follow-Up

For every major review meeting, I use AI to prepare structured agendas based on recent updates, open action items, and known blockers. After the meeting, I generate draft action items from notes. My direct reports have told me this makes our meetings 40% more productive because we spend less time on "what happened" and more time on "what do we do next."

Related: see Stop Drowning in Data: AI Tools for the Street-Smart TPM and Computer Use in Gemini 3.5 Flash: The Agent Actually Clicks the Button Now.

What AI Can't Replace

Here's what's important: AI is terrible at the things that make TPMs valuable. It can't read a room and know that a VP is worried about something they haven't said out loud. It can't build the trust that makes an engineer comfortable saying "I'm stuck." It can't navigate organizational politics or convince two teams with competing priorities to find common ground.

That's the real opportunity — by offloading the mechanical parts of program management to AI, we get to spend more time being the connective tissue that holds complex programs together.

The best TPMs I know aren't the ones who track the most action items. They're the ones who know exactly when to escalate, when to wait, and when to have a hard conversation. AI gives us more bandwidth for all three.