01The setup
Product Management is the new bottleneck.
Engineering is faster. Design is faster. Research is faster. If the PM workflow doesn’t speed up too, the team still slows down — at requirements, testing, prioritization, and decision-making.
Every PM I talk to wants more leverage from AI. They’re using it for the obvious things — meeting summaries, first drafts, polishing emails. It’s helpful, but it’s not yet built into how they work. The gains are scattered. The role still feels like it’s slowing the team down.
This is the playbook for changing that. Three workflows, three levels each, plus a startup-mode ceiling. Each level is incremental. Together they compound — that’s where the 3X comes from. Not faster typing. Removing entire categories of work.
02Workflow one — requirements
Use Jobs to Be Done to give AI the context it needs.
If you ask an LLM to write requirements with no product context, you get generic tickets. If you give it a strong Jobs to Be Done framework — the progress the customer is trying to make, the problem they’re trying to solve, the situation that pushed them to look for something better — you get requirements that are sharper, more grounded, and more useful for engineering.
Keep the JTBD doc to one or two pages. Focused enough that the LLM can actually use it. If you don’t have one yet, dump in whatever you have — customer interviews, support tickets, sales notes, briefs, surveys, rough notes — and iterate. Tell it exactly what you’re creating. Tighten the job statements. Remove anything generic.
One note that matters more than it looks: I use voice for almost all of my writing now. Requirements, rough product thinking, feedback, documentation — talked through first, then structured. AI gets dramatically more useful when you reduce the friction of getting your thinking out of your head.
Level 01 · Structured output
A skill that produces your team’s atomic requirement.
A reusable workflow in Claude Code or Codex: your requirement format, one or two strong examples, your JTBD doc. Goal isn’t faster tickets. Better tickets, faster.
Level 02 · Clarifying questions
Make it interactive.
The skill asks clarifying questions before generating. Ambiguity gets surfaced before it reaches engineering. Fewer rewrites, less back-and-forth.
Level 03 · System integration
Connect it to your actual tools.
Wire the skill to your tracker via MCP. The AI creates the ticket directly — right project, right structure, right status. Idea → voice → structured ticket.
Startup mode · Execution
Don’t just create the ticket — help ship it.
More access, more responsibility. The same workflow that creates the requirement helps implement it. This is where the ceiling is.
JTBD gives AI the context and judgment you normally provide. Skills give you reusable execution. Integration removes the manual overhead.
03Workflow two — testing
QA is becoming the PM’s job.
In a lot of companies the formal QA role is shrinking or gone, and PMs are picking up the slack. This isn’t the glamorous part of the job. Most people don’t become PMs to spend hours clicking through edge cases. But the work matters — and if you let testing eat your calendar, your impact shrinks. The goal isn’t to avoid testing. It’s to make it faster, more structured, and less manual, so you protect quality without becoming the bottleneck.
Level 01 · Manual test planning
Turn acceptance criteria into a test plan.
Have the LLM read the requirement and acceptance criteria, then propose what to test, what edge cases are missing, and the steps to reproduce each. Every test maps back to a specific criterion — so when something fails, you don’t say “it feels broken,” you say “this doesn’t meet this requirement.”
Level 02 · AI-assisted testing
Walk the product, against the spec.
Playwright, browser automation, API scripts driven by Claude Code or Codex. You’re not asking AI to “test the product” — you’re giving it the requirement and acceptance criteria and asking it to test against those expectations. Still review what it flags. It will miss things and over-flag things.
Level 03 · System integration
Failed tests become structured bug tickets.
Requirement, acceptance criteria, expected vs actual, repro steps — AI carries that context forward into a bug in the right place, with the right context. Requirement → AI test → ticket.
Startup mode · Fix the bug
Don’t just file it — help fix it.
You need access to the repo and good judgment about what should and shouldn’t be changed. Validation still matters. But the PM moves from filing the bug to helping push the fix forward.
If testing fills your calendar, your impact shrinks. The fix isn’t to skip quality — it’s to make quality cheaper to protect.
04Workflow three — feedback loop
A tight feedback loop is how judgment compounds.
Once a feature ships, the work isn’t done. You need to know whether it actually worked. KPIs, analytics, support tickets, retention, revenue, customer feedback — whatever your team uses to measure performance. This is where judgment gets built. The only way to develop strong product judgment is the loop: ship, measure, compare to what you expected, learn. If you skip it, you’re shipping without learning — one of the fastest ways to stay busy without getting better.
Level 01 · Manual review
Did the feature do what we expected?
Look at the metrics tied to the feature. Compare what you thought would happen with what actually happened. Run a postmortem — even a personal one — against the original requirement, the tradeoffs, the result. Capture follow-up improvements in the backlog.
Level 02 · AI-assisted analysis
Synthesize across noisy signal.
Give the LLM the requirement, the JTBD context, the acceptance criteria, the post-launch results, and the customer feedback. Ask it to evaluate, summarize, surface patterns. Multiple sources collapse into a clearer picture — but the “so what” is still yours.
Level 03 · System integration
Connect the loop to your product systems.
Pull the original requirement, post-launch results, customer feedback, related backlog items. AI summarizes what happened, identifies gaps, and proposes follow-ups tied back to the JTBD. Ship → collect → synthesize → next actions.
Startup mode · Close the loop
If the fix is small and safe, ship it.
Adjust copy. Tighten an empty state. Update an event. Tweak UI behavior. Compress the gap between learning and improving from days to minutes — with the right guardrails.
The average PM ships features. The high-impact PM ships and learns — and every future decision gets better.
05Where this lands at speed
In startup mode, the three workflows fuse into one.
There’s a fair argument — Lenny’s podcast made it — that you don’t want to be the 51st-best engineer at a company that has 50. Agreed, for large and mid-size. But when you’re the 4th-best engineer at a company that has 3, the calculus is different. At a startup, the same workflow that drafts the spec runs the test, files the bug, ships the fix, and reads the result. The role compresses to a single loop.
The old loop
thinking → writing → revising → coordinating → manual testing → bug write-up → ticket → wait → ship → maybe check metrics later
→
The 3X loop
thinking → speaking → structured ticket → AI test → structured bug → fix → ship → measure → synthesize → learn
The old loop is sequential and lossy. The 3X loop carries context forward at every handoff — that’s where the leverage actually lives.
06Takeaways
2X keeps you in the game. 3X gets you ahead.
This isn’t about AI magically making you a better PM. It’s about removing friction from the work that slows PMs down — writing requirements, validating quality, and learning from what actually happens after launch.
- 01Product Management is the new bottleneck. Speed up the role, or slow the team down.
- 02JTBD gives AI the context it needs to write requirements that are actually useful.
- 03Build a skill for atomic requirements. Add clarifying questions. Wire it to your tracker.
- 04Acceptance criteria become the foundation for AI-driven test plans and structured bugs.
- 05Don’t skip the feedback loop. Judgment compounds; shipping without learning doesn’t.
- 06In startup mode, the same loop that writes the spec helps fix the bug and ship the change.
- 072X keeps you in the game. 3X is how you become impossible to ignore.
Stop being the bottleneck. Become the loop.
