Topic · 105 pieces

AI Product Management

AI did not make product management harder. It collapsed the cost of the parts PMs were trained to be good at, and left the parts nobody trained for.

The role changed underneath the people doing it. Producing options got cheap, so selection became the job; writing the spec got cheap, so the prototype became the spec; hearing customers got cheap, so the bottleneck moved to deciding what you heard means. These posts trace that shift one artifact at a time. Start with the PRD piece if you want the argument, start with the prototyping piece if you want something to do this week.

Start here
  1. 01Salesforce Put a $50 Seat Next to the Credit on Agentforce Coworker. The Seat Is the Budget's Unit.Salesforce added a $50 per-user price to Agentforce Coworker at 185,000 users and 21x usage. The release says why: budget. It does not say what the seat buys.8 min read
  2. 02Nielsen Says Treat a Model Change as a Redesign. The Redesign Your Users Feel Is the Harness Commit.Jakob Nielsen's guideline: treat model changes as redesigns. The change users feel is the prompt cut on a Thursday. My harness: 51 commits, 30 with no name.8 min read
  3. 0384,000 Chats Show When Users Re-Ask the AI. Design Should Own the Repair Rate.A study of 84,000 ChatGPT and Gemini chats counts when users re-ask. When you can't predict the output, design the states and own the repair rate.9 min read

In the handbook

Everything on this topic

105 pieces
  1. 01Salesforce Put a $50 Seat Next to the Credit on Agentforce Coworker. The Seat Is the Budget's Unit.Salesforce added a $50 per-user price to Agentforce Coworker at 185,000 users and 21x usage. The release says why: budget. It does not say what the seat buys.8 min readNov 26
  2. 02Nielsen Says Treat a Model Change as a Redesign. The Redesign Your Users Feel Is the Harness Commit.Jakob Nielsen's guideline: treat model changes as redesigns. The change users feel is the prompt cut on a Thursday. My harness: 51 commits, 30 with no name.8 min readNov 26
  3. 0384,000 Chats Show When Users Re-Ask the AI. Design Should Own the Repair Rate.A study of 84,000 ChatGPT and Gemini chats counts when users re-ask. When you can't predict the output, design the states and own the repair rate.9 min readNov 26
  4. 04Legora Cut the FDE Seat in Two. The Eval Fell in the Gap.Legora has about 100 legal engineers who ship no code and FDEs who never see a client file. Behind a data wall the field owns the rows; only the rate crosses.9 min readNov 26
  5. 05Five Real Users Beat Fifty Synthetic Ones, If You Ask Them How It FeltNielsen: five real users beat 50 synthetic ones, fake users make fake data. My twin had every stuck point right. What it missed, only a question held.7 min readOct 26
  6. 06Reddit's Handoff Ships a Draft PR. Design Should Hand Engineering the Tests.Reddit's feature capsule hands engineering code that may or may not merge. When prototypes are production-shaped, design hands engineering test rows, not code.9 min readOct 26
  7. 07Reddit Has 700 Prototypes on a Dashboard. The Column It Needs Is Shipped.Blair Metcalf showed Reddit Proto on How I AI: 700 prototypes, 100% adoption, real data, and no shipped column. Figma got 17x the code and 1.3x the releases.9 min readOct 26
  8. 08Your Prototype Made Decisions Nobody Made. Who Owns the Spec Now?Ravi Mehta found an AI PRD full of decisions nobody made. When the prototype is free, the PM owns the question, design owns the behavior, nobody owns the rest.8 min readOct 26
  9. 09What Leaves With the Engineer: Get a Copy of the PacketMilos Mandic says buyers should ask what leaves with the forward deployed engineer. Do not ask. Get the handback packet, at two weeks, go-live, and day 60.11 min readOct 26
  10. 10AI Is Startware, Says Nielsen. It Finishes the Day the Failure Inventory Exists.443 designers: 72% use AI in Discover, 41% in Deliver. Nielsen says wait for finishware by 2028. Late errors ship because nobody inventoried them. Ten rows do.8 min readOct 26
  11. 11Buying From an FDE Company: Write the Move-Out Into the ContractA CIO wrote 'stop sending me engineers.' An exit clause gets you out of an FDE contract. Six contract lines get you what you paid for.11 min readOct 26
  12. 12Kill the Correction Queue: The Agent Lands When the Fix Stops Waiting for Its BuilderMarco Di Santo built an agent fast and became its bottleneck. Kill the queue where every fix waits for the builder. Three replacements, and a weekly ledger.9 min readOct 26
  13. 13Salesforce Just Bought the SignalSalesforce is buying Braze for $5.66 billion. Marketing Cloud was built on a send. Braze is built on a signal. The integration is brutal. That's the point.6 min readOct 26
  14. 1417x the Code, 1.3x the Releases. Design's Job Is the Spec Now.Figma shipped a spec editor for agents as its design chief reported 17x more code and 1.3x more releases. Design's job moved from screens to the spec.9 min readOct 26
  15. 15FDE Comp and the Equity Problem: Whose Stock Goes Up When the Loop ClosesIndian FDE postings grew 19-fold at about ₹12 lakh base while frontier labs pay up to 70 percent in equity. The seat is spreading where the upside isn't.11 min readOct 26
  16. 16The Three-Call Theme Was the Renewal. Count What Landed, Not FixesNielsen says AI codes user comments as well as researchers and the test is fixfluence. My ranker buried a 3-call theme under a 31-call one. It was the renewal.8 min readOct 26
  17. 1741 Brands Price AI Five Ways. The Credit Is the New Seat.Only 4 of 41 brands charge per outcome. 21 sell a seat plus credits, and the other columns are drifting there. Who sets the exchange rate is the real price.15 min readOct 26
  18. 18Kill the Finish-Line Price. A Relay Has a Definition Per Handoff.a16z wants agents paid on the settlement at the end of a months-long journey. Fin's one-outcome meter split. My one definition drew 200 disputes a week.9 min readOct 26
  19. 19The Pilot Was Approved on the SixtyA course lesson prices a 120-person claims team at $60 a month, then shows three sensible decisions making it $2,106. Put four cost lines on the handshake.8 min readOct 26
  20. 20The Career Out of the Seat: Draw the Exits, Not Just the LadderThe FDE career ladder is unsettled, and founders are now walking into the seat for paid work. Three exits to write into the offer, and one rule for the ladder.10 min readOct 26
  21. 21Find Your Strikethrough: The Agent Shortlist PanelJason Lemkin's agent picked Render and struck Vercel off in one phrase. No lost deal in any CRM. A biweekly panel of workload questions finds your phrase.8 min readOct 26
  22. 22Five Sessions, Then the Prompting Habit SetsUsers settle on 2 to 4 prompt templates within about 5 sessions. My onboarding agent lowered activation. Put a person in the first week instead.7 min readOct 26
  23. 23The FDE Scorecard: Count What Stayed, Not Who Was SentAnthropic will train 10,000 FDEs and AWS will credential them. Neither tells you whether an engagement worked. Four numbers that do, and five to stop counting.10 min readOct 26
  24. 24Every PM an FDE? The Rotation Is Not the LoopAtlassian has 46 FDEs and says every PM should be one, by mandate. A PM back from one customer is an anecdote. Make the mandate a decision owed to the field.7 min readOct 26
  25. 25Month Two: The FDE Leaves, the Workflow Has to StayGartner predicts 70% of enterprises will abandon FDE-built agentic AI by 2028. The FDE to customer success handoff is where that starts. Here is the page.10 min readOct 26
  26. 26Jev Is 400x Cheaper Per Call. Price the Wrong Pick.Three creators explained Jev by price per call this week. My classifier went from 89% to 61% in six weeks. Rank branches by what a wrong pick costs to undo.7 min readOct 26
  27. 27Agents Don't Answer Interviews. Amplitude's Left a Note.Amplitude added one optional parameter to every MCP tool: why are you calling this. Agents fill it in three quarters of the time. That is the interview.6 min readOct 26
  28. 28The Twin Would Have Shipped ItSalesforce is buying Listen Labs to simulate what customers will do. A twin of my users had every stuck point right and missed how being watched felt.6 min readSep 26
  29. 29In the Room, Not on the Quota: Who Owns the FDE PromiseA Forbes piece put FDEs on the sales team. Job data says zero percent carry a quota. Who owns the promise, and the one-page handshake that says so.10 min readSep 26
  30. 30Which Slice Holds the 3%?Marily Nika asks what is inside a 3% hallucination rate. Mine sat in one slice, a fifth of traffic, and an 89% was 61% by week six. Ask which slice, and when.7 min readSep 26
  31. 31Count the Kills, Not the PrototypesStripe's Harbor has made 12,000 prototypes since May. The number to report is how many were killed in 20 minutes. In my fleet that third paid for the tool.6 min readSep 26
  32. 32Same Model, Twice the Bill: The FDE Codebase ContractA UC Berkeley study found the harness, not the model, sets what an answer costs. So what an FDE changes on site is a margin call. The contract, in three layers.11 min readSep 26
  33. 33Absorb Pain, Excrete Product: The FDE HandbackPalantir made Fujitsu a Global FDE Partner. When someone else does the deploying, the handback has to be a written packet with a date, and a decision back.10 min readSep 26
  34. 34What 267 Posts Are For: A Map of the Field Log267 posts, 49 handbook chapters, 203 answers, 115 templates. What each part is for, who it was written for, and where to start from your seat.10 min readSep 26
  35. 35The FDE Is the Product Builder Wearing a Visitor BadgeOpenAI raised $4 billion for a deployment company. Anthropic launched a $1.5 billion one. The role they hire for is the Product Builder, from the services side.7 min readSep 26
  36. 36Cagan's Definition Is Right. The Weights Are Wrong.Cagan maps Benedict Evans's three product skills onto his risk framework. The skills are right. Their AI exposure is wildly different, and that is the story.8 min readSep 26
  37. 37Polish Stopped Being EvidenceFidelity used to be a proxy for conviction in design review. AI broke that proxy, so the work that looks most finished now gets the shallowest critique.7 min readSep 26
  38. 38The Week the Prototype Beat the SpecA twelve-page spec we argued over for three weeks lost to a prototype I built in an afternoon. The number that ended it, and what I stopped doing.7 min readAug 26
  39. 39What Landing Actually IsShipping is not landing. Landing is when a customer's behavior durably changes and the value sticks. A working definition, and the four ways it fails.7 min readAug 26
  40. 40What Replaces the PM Ladder When the Deliverables Are GoneIf documents were the evidence promotions ran on, and AI collapsed them, leveling needs a new currency. What counts as evidence from L4 to Principal.7 min readAug 26
  41. 41Twelve Builders, Twelve Products, Zero LandingsThe 2026 CPO Insights Report says product builders replace PMs by 2030. The role is real. The framing points at building, just as building stopped binding.7 min readAug 26
  42. 42How I Build Eval RubricsThe eval rubric method I actually use: grade thirty outputs first, tie every dimension to downside exposure, and go binary wherever you can.7 min readJul 26
  43. 43AI Can't Understand the User. That User Is Disappearing.The best defense of the human PM is that AI can't understand the user. But that user is a departmental role, and agents are dissolving the departments.8 min readJul 26
  44. 44Is This Product Bet Group Stage or Knockout? Reversible vs. IrreversibleThe World Cup is two tournaments bolted together. Knowing whether a product decision is group stage or knockout is the same question as downside exposure.5 min readJul 26
  45. 45Stage Wins or the Yellow Jersey: You Don't Win the Tour by Winning StagesThe rider with the most stage wins almost never wins the Tour. A model for AI product management: cumulative outcome over heroics, on a team you never see.4 min readJul 26
  46. 46The Product Builder Was Never About CodingLinkedIn replaced its APM program with Product Builder and framed it as generalists. The real shift is collapsing the distance between deciding and making.3 min readJul 26
  47. 47The PRD Was Already Dead. Slack Proved It in Public.Slack is running tiny cross-functional squads that prototype continuously instead of committing to fixed roadmaps. The org-chart version of the PRD collapse.3 min readJul 26
  48. 48Stop Picking Winners. Design the Tournament.Cheap prototyping did not kill product judgment. It moved the decision from picking the winner to seeding the bracket of what gets prototyped at all.4 min readJul 26
  49. 49Rank AI Features by Downside Exposure, Not EngagementEngagement rewards features people fiddle with. Rank AI features by downside exposure instead: how fast a silent 20% quality drop would cost you customers.4 min readJul 26
  50. 50Storytelling for PMs: The Skill That Decides Whose Evidence WinsStorytelling for PMs is not polish, it is how decisions get made. Three frameworks, the retelling test, and a 4-week practice plan to win the room.8 min readJul 26
  51. 51Downside Exposure, Extended: When the Model Is Wrong About YouDownside exposure covered your AI feature failing a user. The second half: when an AI you don't control represents you inaccurately to a buyer.3 min readJul 26
  52. 52The PRD Collapsed Into Three Artifacts (Brief, Prototype, Eval)The PRD collapsed into a one-page brief, a prototype, and an eval. Answering the four objections 21 PMs raised, with copy-paste templates.16 min readJul 26
  53. 53The PR/FAQ Template: Amazon's Working Backwards for Your BetsA complete PR/FAQ template: press release skeleton, external and internal FAQ, the narrative meeting protocol, and when not to use it. Markdown and Word.6 min readJul 26
  54. 54AI Evaluators Are Your New PM AudienceAI systems now decide if your product gets recommended before a human sees it. Product legibility to a model is a new PM deliverable, not SEO reskinned.3 min readJun 26
  55. 55PM Days 61 to 90: Own One Bet and Kill One CommitmentPM days 61 to 90 in a new job: put your name on one falsifiable bet, kill one inherited commitment, and close the loop with a day-90 note to your manager.10 min readJun 26
  56. 56The Decision Log Template: Train Judgment Like a MuscleA decision log template for PMs: schema, worked examples, the weekly 10-minute ritual, and the quarterly calibration review. Markdown and Excel.6 min readJun 26
  57. 57PM Days 31 to 60: Ship One Artifact That CompoundsPM days 31 to 60 in a new job: skip the quick win, ship one compounding artifact. Four options, selection criteria, a worked example, and the failure modes.10 min readJun 26
  58. 58I'm a Company of One Because I Refused to Vibe CodeI crossed ~$1.5M ARR solo in under two months. It only works because I built agentic engineering, not vibe coding. The substrate does the carrying.5 min readJun 26
  59. 59Kill the 'I Built It in a Weekend' FlexSpeed to demo is a vanity metric. The 'I built it in a weekend' flex measures the cheapest, least durable thing. Ask if it was still running the weekend after.4 min readJun 26
  60. 60The Mental Models That Break When Building Gets CheapMost decision frameworks quietly price building as scarce. When AI makes building cheap, four of them invert and start giving the wrong answer.8 min readJun 26
  61. 61The Vibe Coding Crossover Point: Google's New SDLC WhitepaperGoogle's whitepaper puts the vibe coding crossover at 3 to 10x. The part that costs you is the quiet graduation, and it is a CapEx versus OpEx call.7 min readJun 26
  62. 62The Product Strategy Memo Template (Pyramid Principle)A six-page SCQA strategy memo template after Minto: answer first, three grouped arguments with evidence slots, anti-bets, premortem risks, and the ask.5 min readJun 26
  63. 63Your First 30 Days as a PM: Build a Signal Map, Skip the Coffee ChatsFirst 30 days as a PM in a new job: run decision archaeology, build a signal map of where truth enters the building, and write the manager contract.9 min readJun 26
  64. 64PM Skills 2.0 Is Half a Funeral and Half a BlueprintPaweł Huryn's PM Skills 2.0 review: which of the 68 Claude skills to install (red-team-prd, ship-check) and which half embalms a dying PM role.8 min readJun 26
  65. 65Cagan Built the Church. I Argue We Don't Need One.Marty Cagan's product operating model optimizes for a constraint AI erased. The case for product builders over empowered PMs, and where my thesis breaks.9 min readJun 26
  66. 66The Prototype Graveyard: Why Cheap Demos Kill Good IdeasPrototype-first is my thesis, and it has a failure mode no one names. When building is free, the demo replaces the argument and good ideas die in review.6 min readJun 26
  67. 67The PM/PO Split Is Broken. Merging Them Misses the Point.Paweł Huryn is right that splitting PM and Product Owner is broken. The fix isn't merging them. It's killing the Scrum frame the roles exist inside.7 min readMay 26
  68. 68Three Designer Archetypes in 2026. The PM Split Is the Same Shape.Jenny Wen at Claude hires three designer archetypes; the traditional senior isn't one. The same split hits PMs: builder, specialist, prototype-coder.7 min readMay 26
  69. 69The PM Interview Question I'd Refuse to Answer in 2026PM interview red flags 2026: three questions that signal a translator PM role, three that signal a builder PM role, and the polite exit script for a mismatch.8 min readMay 26
  70. 70Ship Story: The Discovery Week I Ran With Three Agents and No CallsAI agent discovery week, no live customer calls. Three agents (triage, theme synthesizer, opportunity ranker), 20 hours total. Stack, inputs, where it broke.8 min readMay 26
  71. 71The 20-Minute Customer Call Triage AgentCustomer call triage agent: 60-minute call to tagged, theme-clustered artifact with a one-line action in 20 minutes. Stack, prompt, eval rubric, batch run.7 min readMay 26
  72. 72PMs Should Vibe Code Internal Tools, Not Customer FeaturesVibe coding with Claude Code is the best new skill for PMs. But shipping vibe-coded features to customers is malpractice. Build internal tools instead.7 min readMay 26
  73. 73Product Management Isn't Dead. It Escaped the Product Team.Aaron Levie called the emerging role 'agent operator.' The cleaner name: Product Builder plus Agent Operator. Same skill, multiplied across every P&L function.9 min readMay 26
  74. 74Six pushbacks I keep hearing on vibe codingSix confusions I hear most when I say vibe coding isn't a strategy: 'demos sell deals,' 'speed beats clarity,' 'evals slow us down.' Here's how I respond.6 min readMay 26
  75. 75Customer Discovery When Your Customer Is an AgentContinuous discovery assumes a human customer. When your customer is an agent, six methods apply: agent telemetry, failure-mode interviews, capability audits.13 min readMay 26
  76. 76Kill the OKR. The Quarterly Cycle Is Now a Liability.The 12-week OKR cycle is now a liability. Replace it with a Rolling Outcome Ledger: 3-7 active bets, 2-4 week decision deadlines, no end-of-quarter scramble.12 min readMay 26
  77. 77Survivorship Bias in AI: Interview the Planes That Didn't Come BackWald's WWII bomber lesson still runs product management. AI worsens the bias, but agents finally make missing data affordable to study. Here's how.8 min readMay 26
  78. 78Kill the Feature Request Queue: Your Backlog Is a SignalYour 847-item backlog rationed scarce engineering time. Three replacements: signal synthesis, prototype-first investigation, and an outcome bet ledger.10 min readMay 26
  79. 79Continuous Discovery Doesn't Scale for AI-Native ProductsTeresa Torres' continuous discovery is the right answer for human-centric SaaS and the wrong answer for agent products. The punctuated discovery alternative.9 min readMay 26
  80. 80Kill the APM Program. Hire, Train, Promote Builders.LinkedIn CPO Tomer Cohen killed the APM program. The 12-week Product Builder apprenticeship replaces it: four prototypes, one production ship, no rotation.10 min readMay 26
  81. 81Against Product Intuition: Why I Stopped Trusting MineWhy I stopped trusting my product intuition after 20 years. It is pattern-matching on a world that no longer exists, and AI rotates the patterns faster.8 min readMay 26
  82. 82Kill the PRD. The Prototype Is the Spec.The PRD was a 2015 solution to a 2015 problem. Three artifacts replace it: a working prototype (built in Claude Code), an eval rubric, and a one-page README.11 min readMay 26
  83. 83200 PM Job Descriptions Reviewed: 98% Hire for a Dead Job200 PM job descriptions audited. 196 could have been written in 2022. The four tells of a dead JD, four tells of a live one, and a before-and-after rewrite.10 min readApr 26
  84. 84pragmatic-pm-ai-skills Review: A Claude Skill That Tightens PM OutputReview of Valentin Brain's pragmatic-pm-ai-skills: a Claude Skill library that makes PM-AI output incentive-aware, evidence-tied, and free of generic advice.8 min readApr 26
  85. 85How PMs Ship Their First Pull Request With AI ToolsFour-level PM PR ladder, PLANNING.md templates, PM PR review skill file, and CTO pitch docs. Built from real coaching engagements across B2B SaaS orgs.16 min readApr 26
  86. 86Your Product KPIs Are Your Job, DailyA product builder without a daily KPI ritual drives with the rear-view mirror. Pick three KPIs, watch them every morning, elevate to OKRs when they slip.11 min readApr 26
  87. 87Mob Prototyping: One Day, One Room, One PrototypeOne day a week, your PM, designer, and engineer build a working prototype together. Cheaper than spec-driven handoffs. Better output than solo prototyping.13 min readApr 26
  88. 88Build a PM Second Brain from Meeting RecordingsPMs forget 90% of meeting context within a week. How to record, transcribe, embed, and query meetings into a searchable knowledge base that compounds.10 min readApr 26
  89. 89The PM Automation Stack: OpenClaw vs Claude vs Everything ElseOpenClaw went viral. Claude Code went deep. n8n went wide. What each tool actually automates for PMs, what's vaporware, and the exact setup that works today.12 min readMar 26
  90. 90How a 2-Hour Prototype Killed a 3-Month ProjectFive customer calls and a two-hour prototype showed our feature design was structurally wrong. How prototype validation saved 600 engineering hours.8 min readMar 26
  91. 91From Zero to Weekly Discovery in 30 DaysWe went from 'we talk to customers quarterly' to 'every PM runs weekly discovery calls.' It wasn't a mandate - it was a system. Here's the 30-day playbook.7 min readMar 26
  92. 92What Happened When We Killed Our Most-Requested Feature200+ customer requests. Sales demanding it. The CEO mentioned it in an all-hands. We killed it anyway - and activation improved 15%. Here's why.8 min readMar 26
  93. 93The PRD Is Dead. Here's What Replaces It.Product Requirements Documents were a necessary evil in a world where building was expensive. That world is gone.3 min readMar 26
  94. 9420 AI Prompts That Replace 20 Hours of PM WorkCopy-paste AI prompts for product managers into Claude or ChatGPT. Tested across discovery, roadmapping, retros, and stakeholder updates. Save 20 hours a week.22 min readMar 26
  95. 95Prompt Logs Are the New Switch InterviewStop scheduling discovery calls to learn what customers want. They are already telling your AI in plain language. Mine the prompt logs instead.7 min readMar 26
  96. 96The AI-Powered Weekly Review: A System for PMs Who ShipA step-by-step weekly review system that uses AI agents to gather data, surface insights, and prepare your week - in 30 minutes instead of 3 hours.11 min readMar 26
  97. 97How to Build a Working Prototype in 60 Minutes (Step-by-Step)A concrete walkthrough of building a real prototype with AI - from problem statement to deployed, testable app. With the exact prompts and workflow.13 min readFeb 26
  98. 98Jobs to Be Done Has a Half-Life Now (And AI Is Shrinking It)JTBD assumes the underlying job is stable. AI breaks that assumption. Why the job your customer hired your product for is decaying faster than your roadmap.7 min readFeb 26
  99. 99The One-Pager That Replaced Our PRDs (Template Included)The exact one-page template we use instead of PRDs. Covers problem, evidence, solution, metrics, and risks - nothing else.12 min readFeb 26
  100. 100Run a 15-Minute Sprint Retro That Actually Improves ThingsAn AI-assisted retro format that gathers data before the meeting, identifies real patterns, and generates action items that stick.16 min readFeb 26
  101. 101Is Mob Programming the Right Approach for Your Team?Mob programming promises collaboration and quality, but at what cost? A look at why fully empowered Scrum teams are almost always the better bet.4 min readNov 24
  102. 102How to Avoid Survivorship Bias in Product ManagementThe British bomber study teaches us that the most critical data is often what we can't see. Here's how survivorship bias shows up in PM and how to fix it.7 min readSep 24
  103. 103You're Only Listening to the SurvivorsSurvivorship bias is quietly wrecking your product decisions. You're building for the users who stayed and ignoring the ones who left. Here's how to fix that.6 min readSep 24
  104. 104Why Users Don't Know What They Want Until You Show ThemFeature requests are reactive. Visionary product development means anticipating needs users can't articulate yet - and showing them what's possible.5 min readSep 24
  105. 105Show, Don't Tell: Why Users Can't Describe What They NeedFeature requests tell you what's broken. Prototypes tell you what's possible. I stopped treating user feedback as a roadmap and started using it as input.5 min readSep 24
All topicsWorking on this inside a product org? How I help