The Smarter AI Gets, the More It Pays to Talk to Customers
Better models can only work inside the instructions they are given, so the founder who gathers unspoken customer knowledge builds the frame that matters.
digitalnative.tech · The Post-Agentic FounderJared Newman, a Daybreak partner who writes the Digital Native newsletter, was trying to solve a hiring problem: how to spot a founder who will become more capable as artificial intelligence improves. His answer, laid out in a recent essay, starts not with code but with conversation.
He calls the new type the post-agentic founder, meaning someone who builds and directs AI agents rather than doing all the work directly. He is not alone in thinking about this. Y Combinator's Startup School, he notes, has begun asking applicants to share a coding-agent session alongside a resume and short answers.
But Newman argues the more useful signal is not whether a founder uses AI today, but whether she will get disproportionately better as the models get better. And his test is simple: look for founders who gravitate toward user research.
These founders, Newman writes, "spend an inordinate amount of time talking (literally) to users." They watch people work, pick up on subtext that software misses, and uncover problems that customers themselves struggle to articulate. Newman calls user research "a fundamentally human endeavor," rooted in what he calls "genuine curiosity about how people actually live in the world."
What they gather is tacit knowledge, meaning the unspoken context that never shows up in written instructions or training data. Newman calls this "the residual that the labs can't fully eliminate."
That leftover background becomes more valuable as the technology improves, he writes, because smarter systems still work only inside the boundaries someone sets for them. Someone still has to construct that frame.
The smarter the machine, the more it pays to leave the keyboard and visit the customer, because that human understanding of customers matters more as models improve.
Questions people ask
What exactly is tacit knowledge?
Tacit knowledge is what people know but do not write down, like how a job really gets done versus how the manual says it should be done. It shows up in habits, shortcuts, hesitations, and workarounds. Because it is rarely recorded, it rarely appears in the data used to train AI.
Why can't AI just do the customer interviews itself?
In principle, an AI can transcribe an interview or summarize support tickets. What can get lost is the subtext: body language, workplace pressures, and problems customers have learned to live with and so never mention. A person in the room can notice what was not said and ask the next question.
Does this mean technical skill no longer matters for founders?
No, but its role changes. Newman is not saying math or coding no longer matter, only that their relative advantage shrinks as models handle more routine technical work. The scarce skill becomes knowing what to build, for whom, and why, which comes from close contact with users.