Ask the Question That Answers the Rest

Shishir Mehrotra is CEO of Superhuman, the AI productivity company behind Grammarly, which he says has over 40 million daily active users and hundreds of millions of dollars in revenue. He ran product at YouTube, founded the document company Coda, and has sat on Spotify's board for more than a decade. He became a first-time CEO at 21, running a Kleiner Perkins-backed startup, and was coached from 2000 onward by Bill Campbell. He has spent roughly 20 years deciding who gets hired and promoted, and his frameworks on career ladders and "eigenquestions" circulate widely among product leaders. These signals are from Shishir Mehrotra's interview with Marina Mogilko on Silicon Valley Girl.

1. The Recruiting Folder. Mehrotra says nearly all of his own career moves started in conversations that had nothing to do with a job. His Spotify board seat began with a paper he wrote called "formats of bundling," which got passed around until someone told Daniel Ek to talk to him. He runs an AI auto-split in Superhuman that sends every job seeker's email to one folder, which he reads and routes to recruiting. The people he wants to meet land in a different pile. His advice: be interesting in public, start projects and write things people might see, because the recruiting funnel is the wrong place to compete.
2. Managers First. For years Mehrotra told people to become excellent executors and then work their way into management. He now gives the opposite advice. Using AI tools puts even junior people in the position of directing work, which is a different skill from writing the code or the blog post yourself. "Learn every AI tool" is the generic advice, he says; learning to manage is the useful one, and what companies are really hiring for is judgment.
3. Practice in the Driveway. Judgment is learnable, Mehrotra argues, but people try to learn it in the highest-stakes setting available. Nobody learns guitar if every practice is a recital, or basketball if every game is broadcast. At work, every meeting is the recital, and saying "I thought that was pretty good" gets you judged. He tells people to build skills like AI-centric design in side projects, with friends, in loops that give fast feedback, so they are good before the stakes rise.
4. The PSHE Ladder. Mehrotra's career framework runs Problem, Solution, How, Execution. Juniors are handed all four and only execute; later they figure out the how, then invent solutions, and at the top they are handed a space and told to name the problem. A leader running Google infrastructure mapped her whole team on this axis against scope and got an S-curve. Early and late in careers, people grow by scope. In the middle the axis flips, a zone she labeled the "trough of disillusionment," where two people with identical scope get judged on how they do the job.
5. The Ladder Moves Up. AI pushes the whole PSHE ladder upward, Mehrotra says, because everyone now has a strong set of executors. AI can sometimes help with the how, such as getting people on board or testing a market, and sometimes invent solutions. It rarely does anything without a good prompt, and the ability to identify which problem to work on stays "almost entirely in the human's head." The job you do today and the job you get promoted for are drifting further apart.
6. Power Drills and Skyscrapers. Mehrotra dislikes the word "replaced" because it assumes a zero-sum game. When power drills arrived, people asked whether handymen were finished; instead society built skyscrapers and construction employment soared. The person who could run a screwdriver fast stopped mattering, and new skills took over. He grants that the change is moving faster than people are used to, and that 18-to-25-year-olds feel it most, but says the answer is to retrain.
7. One Jingle, a Million Ads. In the Mad Men era, an agency spent a year on one ad with one jingle. At Google, Mehrotra watched firms learn they were building millions of ads and start hiring systems thinkers. The original skill shifted, and taste now has to work across a million ads. When Mogilko said Claude can draft stories, he pushed back: you still need someone with great ideas for Claude to multiply into hundreds of variations, and that person must also be a great judge of which of the hundreds work.
8. Eigenquestions. Mehrotra borrowed the term from linear algebra's eigenvectors. An eigenquestion is the one in a set that, once answered, answers the most other questions, and it is often number 6 on the list. In PSHE terms, P is the question, S the answer, H the delivery and E the execution. Skill at eigenquestions became his number one indicator of someone he wants to hire, and his teams train and recruit for it.
9. The Teleportation Interview. Mehrotra has asked hundreds of candidates, from engineers to salespeople, how they would bring a newly invented teleportation device to market. He lets them list every question, then says the scientists are annoyed and they get only 2. One standout drew a 2x2: is it safe for humans, and is it more expensive to buy or to use? Cheap-to-buy units go everywhere, like phones; expensive ones go where airports go and get funded through government; unsafe ones haul trash or ship transplant organs to Africa. He says a 5-year-old can answer it, and keeps a list of about 100 such prompts for low-stakes practice.
10. The AI Superhighway. Grammarly runs over 100 billion LLM queries a week, more than 3,000 per user per day, across about a million unique apps and surfaces daily. Mehrotra says its core technology is bringing AI to where people already work, and until 3 weeks before the interview only one car ran on that highway: "your high school grammar teacher." Superhuman split the product, turning Grammarly into one agent and building Superhuman Go on the infrastructure underneath. Users build agents with their own data, connectors and prompts, and the key trigger is "while writing."
11. Assist Beats Chat. Mehrotra sorts AI products into 3 modes: chat, do (clearing a task list) and assist, which shows up where you work before you ask. A heavy ChatGPT or Claude user might send 10 queries a day; Grammarly users generate thousands because nobody has to remember to ask. His own agents fact-check revenue numbers in emails, fill bracketed placeholders in his drafts, and underline "we should meet tomorrow" with a free slot from both calendars. One magazine publisher with 16 review departments turned all of them into a single assistant that flags problems while reporters write.
12. Bill Campbell's Bar. A year into working with Bill Campbell, Mehrotra offered him advisor equity, and Campbell told him to keep it and grow the company. Asked what motivated him, Campbell said he kept a list of people he had worked with or mentored who became Fortune 500 CEOs; in 2000 it already had 20 names, and he knew them by heart. At Campbell's funeral, Mehrotra says, every speaker told nearly the same story: he had infinite time for them and only cared about their success. The lesson Mehrotra took is to root for your people, even when they leave, and to judge your own success by theirs.

Watch the full video at https://www.youtube.com/watch?v=JEZVl3BkSLk.