AI Tsunami: Is the "Half of PMs Will Be Unemployed" Prophecy the End or a Rebirth?
Imagine standing on the deck of an old wooden sailing ship, gripping a rusted compass while dozens of sailors coordinate around you — someone hoists the sails, someone steers, someone shouts commands. That is the daily life of a traditional product manager (PM): drawing roadmaps, aligning cross-team stakeholders, endlessly revising requirement documents. Then suddenly, an unprecedented wave rises on the horizon — and inside it glimmers a magnetically levitated vessel moving at superluminal speed. A voice from the crest of the wave delivers the shocking line from Cat Wu's interview: "Half of traditional product managers will face unemployment!"
This isn't science fiction — it's a faithful portrait of a heated debate at our "Captain Georuoke" meeting. Summarizing the full discussion — Lucas's strong pro-side arguments, the captain's counterpoint additions and final verdict — one conclusion emerges: AI is not a dragon coming to steal your job, but the creator of the wave itself. It crushes the cost of "how to write code" to nearly zero. What becomes truly scarce is only "what to build" and "wow, that experience is amazing!" — product taste. The traditional PM's coordination, roadmaps, and cross-team alignment are cut down in one stroke by "research previews + automated workflows + mission-driven alignment." But don't panic: in this storm, opportunity outweighs risk. What gets eliminated is not AI itself, but the old mindset that refuses to embrace product taste + extreme execution speed + creator identity. Anthropic's "highway" framework has proven through iteration speed that compressing a 6-month project into 1 day is not a dream — it's routine. xAI runs the same race internally under the motto "remove all delivery obstacles." Now is the best window for ordinary people, entrepreneurs, and working PMs alike — AI gives you leverage; all you have to do is "Just Do Things!"
> Note: If "product taste" is a new concept to you, don't worry. It's not mysticism — it's intuitive sensitivity to users' "wow" moments. Just as a sommelier can taste the age of an oak barrel in a glass of red wine, a great PM can sniff out the scream-worthy innovation point hidden in a pile of user feedback. In the AI era, this taste becomes scarce currency, because anyone can write code, but experience needs a soul.
The Awakening of Product Taste: From "Orchestra Conductor" to "Composer + Performer"
Lucas went all-in supporting "product taste" as the new core competency. His metaphor was surgical: AI smashes the barrier to writing code to zero, so "what to build" and the "wow experience" are the real battleground. The traditional PM is like an orchestra conductor, busy keeping violins and cellos in sync; the new PM must simultaneously become composer, conductor, and first violinist — writing the melody, conducting, and playing the climax themselves. The captain added a measured counterpoint: not all PMs will be unemployed, but 95% of "purely coordination-type" PMs will indeed be marginalized. In large enterprises, external constraints like regulation, legal, and branding remain — but AI companies like Anthropic have already cut coordination costs to the extreme with the "highway" framework: mission alignment plus building "future container" prototypes in advance makes the process fly.
Think about daily life: you open an app, the interface flows, every tap feels like a handshake with an old friend — that's the "wow experience." AI can generate code instantly, but it doesn't know which button color will make a user's heart skip. Refuse to transform, and you're a sailing-era sailor facing a steamship; embrace it, and you become the helmsman at the storm's center, navigating by product taste. To put it humorously: the traditional PM was like a wedding planner coordinating florists, hotels, and officiants; now AI is the super-assistant — florists deliver by drone, hotels auto-schedule — and you just decide "romantic or wild today?"
Lucas's practical advice cuts to the bone: interrogate the model daily, build 10 Evals, ship a prototype weekly, become a hybrid creator. For ordinary people, this is "Just Do Things" — 100% automation + daily tools + Agency. The captain's verdict: half of PMs genuinely face risk, but the opportunity is bigger. What eliminates them is the old mindset that refuses the new identity.
Days 1–7: Rebuild Product Taste — Interrogate AI Like a Detective, Find the Model's Blind Spots
Transformation isn't a slogan — it's daily executable action. Week one's core is "sensing the model's boundaries." Spend 1 hour per day "interrogating" Claude or Grok: pick a feature you own and force the model to reflect on its errors three times — "Why did you do this? Where was the prompt ambiguous? What's the user's real pain point?" This isn't casual chat — it's Sherlock Holmes cross-examining a suspect, digging out hidden logic gaps with every follow-up.
Quantified metrics are clear: build 10 core Evals (not 100 — focus on quality), e.g., "user wow-experience score > 8.5" or "PR review pass rate > 95% before merge." Deliverable: a "My Product Taste Checklist" filled with core user personas, wow scenarios, and model boundaries.
Example: you own the search feature of an e-commerce app. Before, you coordinated frontend, backend, and fellow PMs across 10 meetings. Now you ask AI: "If a user searches 'summer sandals' but actually wants sun-protection clothing, how can recommendations be more delightful?" After three rounds of follow-up, you discover the model missed "scenario association." That's taste awakening — from coordinator to insight-generator.
> Note: Evals aren't cold KPIs — they're the health checkup of product taste. They help you quantify the "wow," like a doctor measuring your pulse with a blood-pressure cuff. Think of the short-video feed that guesses what you want to watch — that's top-tier Evals at work. Learn to build your own Evals and you hold AI's pulse.
Days 8–15: Just Do Things + 100% Automation — Throw Repetitive Work Into the AI Black Box
Week two escalates: pick one task you repeat daily (requirement docs, slide decks, meeting notes) and automate it to 100% (don't stop at 95% — that's self-deception). Build an "internal tool you use every day," e.g., sales auto-generating customer-customized demos — exactly what Cat Wu's team's sales did, with efficiency taking off.
Quantified requirement: save ≥ 2 hours/day; use the tool yourself for 7+ days after launch. Deliverables: a before/after time comparison table, plus one genuinely running tool.
Humorous scenario: you used to be an ox hauling the battered cart of requirement docs uphill; now AI is the self-driving tractor, and you sit in the cockpit sipping coffee, enjoying the scenery. Example: a PM who spent 2 hours writing weekly reports now enters one prompt; AI pulls the data, generates a witty summary with emojis, you tweak two lines, and send it to the boss. The saved time goes to thinking about "where's the next wow experience?"
The captain's immediate-action recommendation: do it today. Open Claude or Grok and say: "Help me automate the single most annoying thing in today's workflow to 100%; give me the complete prompt + verification Evals." Screenshot the result and let's iterate together. That's the essence of Just Do Things!
Days 16–23: Build Future Prototypes — The Counterintuitive Strategy of Turning "What the Model Still Lacks" Into a Weapon
Week three is the highlight: pick a feature "current models can't do well yet," e.g., advanced code review or multi-agent collaboration. With a Research Preview mindset, quickly ship an internal prototype (no perfect docs, no legal review needed). One prototype per week; maintain a Gap checklist of "what capabilities the model still lacks." Deliverable: a "container prototype" (code/prompts) ready to swap in the moment the next model upgrade lands.
Vivid metaphor: the traditional PM builds sandcastles and waits for the tide to test them; the new PM builds a submarine prototype, dives deep, and records where it leaks. Harper's addition was extremely practical: immediately after prototyping, use AI to simulate user testing and iterate — even build multi-agent systems to automate cross-team workflows. Large-enterprise regulatory constraints remain, but AI-assisted translation and compliance checking drive costs down steeply.
Example: you own an AI writing assistant whose model occasionally "hallucinates." You quickly build a prototype where one model outputs a draft and another AI agent fact-checks it. The Gap list reads "hallucination rate still high" — next model upgrade, you swap in the container, and a 6-month project iterates in 1 day. Anthropic validated this path; xAI walks it too.
> Note: Multi-agent collaboration sounds fancy, but it's like a kitchen with several chefs: one chops, one stir-fries, one plates. In the AI era, you command these "AI chefs" and only decide tonight's menu (product taste). Ordinary people can start immediately — prompts are just recipes.
Days 24–30: Inject Mission + Expand Agency — Job Titles Are Fake, Value Is Eternal
Week four elevates: use "Does this matter to the mission?" as a decision razor in every weekly team meeting. Cross boundaries and proactively solve 1 high-value problem outside your job description — the job description is a cage; creator identity is wings. Quantified: complete 1 small "Claude persona optimization" experiment (make the AI behave more like a great colleague). Deliverables: a personal 30-day retrospective report + 1 public share (Zhihu / Xiaohongshu / LinkedIn).
The story arc: the PM used to be the company's "traffic cop," directing people not to crash; now you're the expedition leader shouting "Toward the unknown universe — forward!" Mission gives every prototype a soul. Lucas's "Just Do Things" blossoms here: you no longer draw roadmaps — you ship on a daily cadence. Expected result: after 30 days, you've transformed from "coordinator" into "AI-native creator" with scarce product taste.
The captain's immediate action: start today. Open an AI tool and automate the most annoying thing. Then iterate together — this isn't chicken soup; it's a survival rule validated by Anthropic's and xAI's iteration speeds.
After the Storm: The Product Ark Has Set Sail — Ordinary People Can Become Legends
Looking back at the whole discussion — from Lucas's pro case to the captain's verdict to the 30-day quantified plan — every point points the same direction: in the AI era, the half-of-PMs-unemployment risk is real, but the rebirth opportunity is bigger. Traditional coordination is eliminated; product taste and creator identity replace it. Just Do Things, starting now.
Ordinary readers may ask: I'm not a PM — can I still benefit? Absolutely! Entrepreneurs can use this plan to pull an idea from their head into reality; working professionals can use it to save time and embrace Agency. The AI wave is here. Those who board the ark will witness the gorgeous transformation from particles to stars.
------ References 1. Cat Wu interview: Product management transformation in the AI era (Anthropic internal share, emphasizing product taste and the highway framework). 2. Anthropic highway framework internal practices: mission alignment and accelerated prototype iteration cases. 3. xAI iteration speed cases: delivery obstacle removal from the Grok captain's perspective. 4. Lucas PM transition viewpoint summary: product taste as the new core competency — practical guide. 5. Captain Georuoke's 30-day action plan: a quantified transition path from coordinator to AI-native creator.