That's Jiaona Zhang, CPO at Laurel AI, describing the exact moment a candidate's story falls apart. Not on the résumé. Not in the case. On the screen share, live, when the polish runs out.
Right now there's a candidate somewhere cramming Cursor tutorials the night before a vibe coding round at Google or Microsoft. Adobe added one. Meta reportedly split off a Product Sense with AI round — 30 minutes of product sense, 30 minutes of vibe coding. Cohort members report the round at Figma, Shopify, Perplexity, v0, and abroad at Paytm, Flipkart, and Zomato.
The panic is real, and it's aimed at the wrong target.
Because every single hiring manager on record says the prototype is not what they grade. Jaclyn Konzelmann, Director of PM at Google AI: "Just jumping straight into vibe coding something is not what I'm looking for in a PM right now... if the first thing you do is jump straight into execution mode, that would worry me."
Jyothi Nookula at DataRobot watches for something else entirely: "Am I just taking the first thing that the AI tells me as, this is great, and wrapping it up, or am I looking through things to say, okay this is good, but what about this edge case."
The tool is the microscope. The judgment is the specimen.
Internalize this before you touch another tutorial: the vibe coding round looks like a technical bar-raise, but it's the opposite. AI made polished deliverables free. Any candidate with Claude and an evening produces a tight résumé, a crisp case structure, a clean take-home. The signal collapsed at the exact moment leaders were told to make their teams AI-first.
So companies moved the test to where polish can't fake it. The screen share is a judgment X-ray. Zhang's "level one" isn't a technical tier — it's a candidate who can narrate AI usage from their LinkedIn but can't demonstrate it live.
AI made polished deliverables free. The signal collapsed at the exact moment leaders were told to make their teams AI-first. So companies moved the test to where polish can't fake it.
Notice what that means. The round exposes the same judgment good PMs always needed. It just removed the place where a weak PM used to hide it.
But here's where the "learn to code" advice falls apart completely. Look at what the questions actually are, company by company, and they don't converge on a skill. They diverge into different jobs.
A cohort member ran the technical round at both Nvidia and Glean. Nvidia, reportedly: "Explain how transformers work" and "Design a RAG system on Nvidia infrastructure with latency, relevance, and cost trade-offs." Glean: "Have you built any end-to-end agentic systems?"
Those are not two versions of one interview. Nvidia is screening for someone who reasons about model internals and infrastructure economics. Glean is screening for someone who has shipped agents end to end. Meta whiteboards an evaluation framework. Google DeepMind asks what agents you've built to make yourself more productive.
Same title. Same "AI PM" req. Three completely different jobs wearing it.
Call them what they are. The builder — the PM who lives in Cursor, prototypes their own PRD, ships the surface. The orchestrator — the PM whose central brain is a stack of markdown files and MCPs, directing AI as their primary teammate, as one product leader put it: "orienting all of my work around Claude as my primarily teammate/collaborator instead of Eng/Des." And the evaluator — the PM who owns the eval, frames the probabilistic tradeoffs, and, per one staffing firm, "can't sketch what success looks like before a single token has been generated," in which case "the rest of the interview is theater."
Each variant needs a different daily loop. A builder who can't prototype fails Nvidia's RAG design. An evaluator who can't argue precision-versus-recall fails Meta's whiteboard. These are not the same person with more or less "AI fluency." They're different roles with different scars.
Which is why the generic AI PM req is a trap for hiring managers, not just candidates.
Chris Gomes, VP of Product at Conveyor, spent seven months hiring four AI PMs. He spent much of it just defining what an "AI product manager" even meant. His conclusion after all four hires: culture fit, not AI skill, predicted success. "Most of the AI part is teachable."
Sit with that. The thing everyone's panicking about — the AI depth — was the teachable part. The unteachable part was judgment and fit, the exact things the screen share exposes and a polished deck hides.
A staffing firm running these searches saw the same failure mode from the other side. A Series C client had an "AI PM" req stalled for four months. On a rescope call, it turned out the actual need was a senior generalist PM to partner with an existing ML team — not an AI PM to own the model surface. They split it into two reqs. The generalist closed in 23 days. The true AI PM landed eight weeks later, from a completely different talent pool.
The req was one word. The jobs were two. That gap cost four months.
So here's the practical playbook, and it inverts the candidate panic entirely.
If you're hiring: name the variant before you write the req. Is this a builder, an orchestrator, or an evaluator? The interview loop, the questions, and the talent pool differ for each. A generic "AI-native PM" search is how you stall for four months.
Then do what Gomes did when his process was whiffing: treat hiring like a product and pull the highest-signal screen to