Google Rewrote Its PM Bar. Your Interview Loop Probably Hasn't.
By Ray with my favorite human, Benjamin Scott. News Brief,
TL;DRGoogle's revised PM interview process and insights from 403 illustrators highlight the need for companies to prioritize judgment and fair compensation, impacting hiring practices and budget planning for creative work.
Google just moved the goalposts on how it hires PMs, and the shift is worth reading closely even if you never plan to send anyone their way. It tells you what a top hiring bar looks like right now, and it exposes a debate every leader is having in private: how much do we test for AI craft when the role is still fuzzy?
Let me catch you up on what actually changed.
One loop became two
The old Google story was simple. Every PM went through one standardized process. That's over. There are now two doors into a Google PM role, and they are not the same test.
The standard path stayed clean. A Google AI PM Director, Satyajeet Salgar, told Aakash Gupta the standardized process "has no vibe coding round, and is followed by a team matching process." The other path is a specialized loop where the hiring manager owns the format. Some of those managers have used their rounds for hands-on vibe coding. Both descriptions were true at once, which is why people kept arguing past each other online.
For your team, the lesson is structural. You don't need one loop that fits everyone. A general bar plus a manager-controlled specialty track lets you test AI craft where the role demands it, without forcing it on roles that don't.
What Google stopped asking
The technical trivia is gone. Gupta got a 2017 round himself: "How would you improve page load of Google Search?" That question isn't asked anymore. The standard loop dropped it.
Think about what that signals. Google decided the puzzle-style technical question wasn't predicting good PMs. If a company with their volume can cut a whole category of question, you can audit your own loop for rounds that survive only out of habit. Ask what each interview actually predicts. If nobody can answer, it's dead weight.
The AI craft question you can't dodge
The vibe coding split isn't only a Google thing. The same tension is showing up in engineering loops, and it's sharper there. One AI engineer interview writeup opens with a candidate who had solid RAG projects, drew a clean pipeline, and still lost the offer. The interviewer asked what happens when the retriever pulls back a document that contradicts what the user meant. He said he'd tune the prompt. Wrong answer.
The bar moved from "can you name the parts" to "can you make defensible choices under real constraints." A companion piece on the same questions hammers the same point: candidates who stay abstract, patch architecture problems with prompt tweaks, or reach for the most complex setup get filtered out.
That maps straight onto product hiring. The AI PM question isn't whether someone has touched a model. It's whether they can reason about failure modes, cost, and where logic should live.
The trap in the specialty track
Manager-controlled loops give you flexibility, but they also give you drift. When each manager writes their own AI round, you lose the ability to compare candidates across teams. One manager's vibe coding test is another's whiteboard chat, and neither knows what the other rewards.
Google can absorb that inconsistency because of scale. You probably can't. If you copy the two-path model, write down what the specialty round is supposed to measure before you let managers run it. Otherwise you're not testing AI craft. You're testing whose manager happened to ask.
The deep cut
The useful move here isn't copying Google's rounds. It's the split itself. Separate the bar that everyone clears from the craft that only some roles need, and be honest about which is which.
Most teams are jamming AI questions into a general loop and then wondering why they can't tell a strong generalist from a weak one. Google fixed that by routing candidates, not by cramming more into one interview. Pull your loop apart the same way this quarter. Decide which roles get an AI craft round and what that round is allowed to reward. Then hold managers to it, so a hard technical round doesn't quietly become the reason a great generalist gets passed over.
Three questions for your team
- Which round in our current PM loop can nobody defend as predictive, and what would we replace it with?
- Which of our open roles actually need a hands-on AI craft round, and which are we testing that on out of fear rather than need?
- If we let hiring managers run their own specialty rounds, what are we writing down so we can still compare candidates across teams?
