Human.
Everything downstream depends on this layer — and it's the one most orgs are quietly hollowing out.
It's the instinct to interrogate a number before trusting it, but also to know which bet is worth making, which feature is a distraction, what a customer means versus what they said. To read a metric, a roadmap, and a user interview and see the same problem underneath all three. That judgment isn't a tool you install. It's pattern recognition built over years of leading product across data, design, and strategy — of being close enough to the work to see what dashboards and frameworks can't.
I still do that work directly — the segment models and regressions when the numbers don't add up, but also the positioning calls, the prioritization tradeoffs, the customer conversations that change a roadmap. Not because I have to, but because it's where the judgment stays sharp. And that judgment is the scarce input now. You can't encode what you don't understand. You can't teach an agent to weigh a tradeoff you've never learned to make yourself. The systems and the agents are only ever as good as the product judgment underneath them — which makes that judgment more valuable in an AI-native world, not less.
You can't encode judgment you don't have. This is where it comes from.
System.
Then I encode that judgment so it runs without me.
Customer signal is everywhere; most companies don't have a signal problem, they have a system problem. Signal arrives in fragments, gets discussed once, and disappears into a Slack thread nobody can find again. So I build the pipeline that catches it: NPS themes on rolling 14, 30, and 60-day windows, qualitative signal fused with behavioral data, a weekly digest that lands without anyone pulling it. The short window catches what's emerging; the long one shows what's structural.
The judgment is in the design — which signals matter, which windows reveal what. That's the part a dashboard can't do for you. And it compounds: pricing research I ran in year one was still authoritative five years later. The system still runs, unmaintained. I build things that outlast my involvement.
If your roadmap has twelve priorities, you have no priorities. Synthesis is the art of the hard no.
Agent.
Then I hand that system to agents that run it while I sleep.
I built an autonomous PM agent — Sasha — in about ten minutes, by dictating what I wanted to Claude. She runs every night: monitors Slack for tagged requests, researches the competitive landscape, drafts tickets, generates mockups. She handles the no-brainer features that should ship but always get crowded out.
The ten minutes is the punchline. The decade of judgment that made it possible is the story — Sasha is only as good as the patterns I taught her to look for. That's the part most people miss: an agent doesn't replace the experienced operator, it scales them.
If your team can only ship when everyone is in the room, you don't have a system. You have a dependency graph shaped like a person.