Hadar Zeharia

Senior Product Manager

I build what others
build on top of.

A decade in platform work — Stripe Payment Links, Masterschool through its $100M seed and twenty-country build-out, Embedded Finance at eOS. The work that compounds is the work underneath the screens: the APIs, the ledgers, the state machines, the contract between the platform and the operators on top of it. This site is the field notes from that layer.

Portrait of Hadar Zeharia
BuildingPM OS · T4 boundary calibration×ShippingEmbedded Finance at eOS · Israel×ReadingHow to Take Smart Notes · Ahrens×ListeningAcquired×Watchingagentic commerce · stablecoins×Draftingaudit-log-as-PM-artifact×Wonderingwhat AI judgment looks like at company scale×Open filehadar.pm×BuildingPM OS · T4 boundary calibration×ShippingEmbedded Finance at eOS · Israel×ReadingHow to Take Smart Notes · Ahrens×ListeningAcquired×Watchingagentic commerce · stablecoins×Draftingaudit-log-as-PM-artifact×Wonderingwhat AI judgment looks like at company scale×Open filehadar.pm×

AI-augmented · autonomous PM

The engine proposes. I choose where to go.

Most PMs use AI as a chatbot. I designed and built an autonomous PM operating system — 19 agents that listen across my channels, draft strategy, propose roadmaps, surface convergences. Every morning, takes and drafts reach my desk. The judgment — what to actually ship, what to overrule, where to compound — stays mine.

The architectureCowork
  1. Judgment

    Me. The only loop AI doesn't close.

  2. Reflection2 agents

    A retrospective agent (weekly scoring) and a meta-agent (monthly prompt tuning) — the system improving itself.

  3. Synthesis4 agents

    Pattern detection · strategy drafting · culture learning · decision drafting.

  4. Listening12 agents

    Scouts on cron — GitHub, Jira, Slack, Gmail, calendar, competitors, regulators, customer voice.

  5. Foundation1 agent

    Cowork runtime, 6 MCPs, scheduler, audit log, plus the action executor that runs Tier 4 actions after `/pm-approve`.

What reaches my desk every morning

  • Strategy takesDrafted opinions on convergences across 7-day windows, ready to approve or overrule.
  • Roadmap proposalsSpecific bets the synthesis layer thinks deserve compounding — with the alternative it rejected.
  • Drafted memosBriefings, decision memos, ticket drafts — sitting in drafts until I sign or rewrite.

0 agents propose.
1 me decides.

The premise is simple. AI is good at listening, drafting, and detecting patterns. I’m good at judgment. The architecture separates those — and the autonomy model makes sure the line never blurs.

Graduated autonomy · 5 tiers

  1. T1Read anythingauto
  2. T2Write to memory / Obsidianauto, audited
  3. T3Draft artifactsauto, lands in drafts
  4. T4Create Jira / Confluence / Slack/pm-approve
  5. T5External comms · deployhuman only
Read the full case study

What I work by

Four claims worth arguing about.

Each one is a position, not a slogan. If you disagree with any of them, that's the conversation worth having.

  • 01

    AI doesn’t take the PM job. It takes the parts you didn’t actually want.

    Listening, drafting, pattern-detection — agents are great at all of them. Judgment isn’t a part of the job. Judgment is the job. And it doesn’t shrink as the engine gets stronger. It gets concentrated.

  • 02

    The product is the contract underneath.

    Screens are the cheapest part. The contract underneath is the part that compounds — the spec, the state machine, the agreement between a platform and its regulator. Most product debates are contract debates dressed up as feature debates.

  • 03

    The org chart is your second product.

    Every product is, in the long run, a function of the org that ships it. The features matter. The team and the operating cadence around them matter more. Founding seats teach you this fast.

  • 04

    Make the system legible before you change it.

    The most underrated PM job is describing what already exists — accurately, currently, in one place everybody points at. The legibility work doesn’t show on a quarterly OKR. It shows in the speed of the team six months later.

Career path · interactive

Law to API contracts.

Click any node for the take. The story doesn't go in a straight line; few do.

About

Who's writing this.

The thing I liked about law school was reading contracts. The *systems* part — who owes what to whom, where the contract holds and where it leaks. Product is the closest job I've found that asks me to do exactly that for a living.

Stripe Payment Links: the contract was the API. Masterschool: the deal between a learner and a school six months in the future. eOS now: between a banking platform and the bank running on top of it. Different rooms. Same job.

I built PM OS because the parts of the work I actually want to do are getting smaller and more concentrated. Nineteen agents on Cowork listen, draft, and surface convergences across my week. They produce. I judge.

Tel Aviv. If you're building something where the contract underneath is the interesting part, I'd like to hear about it.

Now · May 2026

What I'm working on this quarter.

Inspired by Derek Sivers’ /now page. Updated quarterly.

  • Embedded Finance at eOS. Public-facing detail in the case study. The rest is NDA.
  • PM OS in production every working day. Currently calibrating the T3 → T4 boundary as the synthesis layer's draft quality climbs.
  • *Lords of Finance* (Ahamed) at night. The 2024 Synapse trustee filings when I forget how sponsor-bank composition actually fails. *Complex Systems* (Patrick McKenzie) on the walk home.
  • Sharpest open question on my desk: where exactly the line lands between PM judgment and AI judgment over the next 24 months.

Contact

Say hello.

Email is the best signal. I read every message.