AI Means You Need Your First Product Manager Sooner
Faster engineers create more decisions to make, and unpredictable models add chaos, so the old wait until six to eight engineers shrinks to four or five.
First Round · When Should You Stop Clinging to Product and Hire Your First PM? Advice From a Founder and CPOBrandon Levey, founder and chief executive of Ichi, an AI code-compliance tool, was explaining to Saumil Mehta, a product leader who wrote the piece, why his staffing math had changed. “Our engineers are now probably twice as productive as I’d have expected a few years ago because of AI.”
Mehta argues that Levey's experience points to the opposite of the popular fear that AI kills the product manager. The productivity boom from AI, he believes, will on balance push startups to hire their first product manager sooner than they would have before AI.
The reason, as Mehta tells it, is a flywheel. Features now ship much faster, which creates more features to ship, more decisions to make, more inputs arriving faster, and more judgment about what is actually good. Product managers, he argues, are especially well placed to do that sorting and choosing.
Stanford machine learning professor Andrew Ng made the complements argument in a written piece, describing product managers and software developers as complements, like cars and gasoline, which supports Mehta's view. When coding gets cheaper and faster, teams need more product management as a share of the workforce, not less. Levey, a second-time founder who had led platform product as a general manager at Square, is living that shift. Levey said, “Our designer is now also able to double as a frontend engineer,” and added, “I’m thinking about reducing the ratio between eng and PM and adding one more PM.”
The benchmark moves with it. Mehta says startups once waited until they had six to eight engineers, on the way to the common average of one product manager for each eight to ten engineers. Now, with much more productive engineers and designers, he says the time may come after bringing on just four or five engineers. The caveat is that a lot of the rote product work in documentation, program management, and status updates will be automated, and at larger size, companies may employ fewer product managers than they once would have.
The second push comes when the startup itself is building with AI. Levey told Mehta, “There’s a lot more chaos in the system compared to general software being built.” That unpredictability in how the models respond, Mehta argues, spills into product planning, customer research, and small iterations, and creates new product work that did not exist before.
Questions people ask
What does a first product manager actually do day to day?
A first product manager turns founder guesses, customer requests, engineer concerns, design ideas, and usage data into a clear order of what to build next. That means running sprint planning, triaging bugs and feature requests, writing clear specs, and saying no to good ideas so engineers can ship the best ones.
Why are AI responses called non-deterministic?
Large language models generate answers by predicting likely next words, so the same prompt can produce different wording, tone, or even facts on different tries. Small changes in phrasing, context, or a new model version can also change the output, which makes testing, quality control, and product planning harder.
Are engineers really twice as productive with AI?
Estimates vary widely, but improvements of roughly 20% to 50% are often cited for routine coding tasks with AI assistants, with wide variation by task and engineer skill. Levey's doubling reflects his own team's experience, not a universal benchmark.