AINEWS
Notion Stopped Hiring for Experience. It's Hiring for Curiosity.
Inside Notion, an AI coding agent is changing who ships code, who gets hired, and what seniority means on a software team.
Inside Notion, an AI coding agent is changing who ships code, who gets hired, and what seniority means on a software team.
At Notion, an engineering org chart built on tenure is being rewritten around an agent that can return a shippable first cut from a feature request.
The shift shows up in two places at once. Managers who had not shipped production code in years are back in the codebase, working alongside the engineers they lead. And the hiring bar has moved from "years of experience" to "curiosity and open-mindedness," because, as Ryan Nystrom, who runs AI Product Engineering at Notion, puts it, the experience the field normally calls for "doesn't exist yet." Both observations come from Nystrom in OpenAI's customer story "What Codex unlocks for Notion", a vendor-published case study that should be read as a single-source testimonial rather than an independent audit.
Nystrom's framing is blunt about how much of the old model has broken. Engineers at Notion now spend more time writing spec documents than writing code by hand, then hand those specs to Codex and let the agent return a first cut. "Honestly, I don't really write code by hand anymore," Nystrom told OpenAI. The work that used to scale by adding more bodies is scaling, instead, by adding more well-scoped tasks for the agent to run in parallel overnight.
The concrete example Nystrom offers is a voice-input feature on the web. The mobile version already existed; for the web, the team handed Codex a spec, the existing mobile implementation, and a verification hook, and the agent produced a first cut that, in Nystrom's account, "matched Notion's codebase conventions closely enough to ship the next day." Nystrom frames the cycle time as two weeks collapsed to three hours, but the two-week baseline is his own self-estimate of what the work would have taken, not an externally measured benchmark. It is a useful data point about how one engineer thinks about the work, not a productivity study.
That is also where the org-shape claims start to matter more than the speedup. If an agent can absorb the first cut of a feature, the leverage point inside a software team moves away from typing speed and pattern recall, and toward the quality of the spec, the verification hook, and the judgment about whether the cut is good enough to ship. Senior engineers still matter, but the part of seniority that consisted of being the fastest typist in the room is no longer the bottleneck. Notion's response, on Nystrom's account, has been to push more of the org into the codebase rather than pulling more of it out. Managers who had drifted into review-and-roadmap work are pulling changes again.
The hiring change follows the same logic. In a market where the experience that predicts success at agent-augmented engineering cannot yet be measured in years, Notion is hiring for curiosity, on Nystrom's read, because the field's traditional signals are a year or two old at most. That is a defensible position when you are one company rewriting its hiring rubric. It is a much harder position to defend as a general rule, and Nystrom does not try. The OpenAI piece is a customer story, not a labor-market study. Whether "curiosity" holds up as a hiring signal at scale, and how much of the speedup depends on a single motivated engineer plus a willing vendor tool, are questions the case study does not address.
The narrower, more durable claim is the one Notion is actually making about its own engineering culture. The company is treating agent-augmented coding as a structural change, not a productivity hack. That is why it is rebuilding primitives and abstractions so that agents can use them, why it is pulling managers back into the codebase, and why it is dropping years of experience as a hiring filter. The vendor headline frames this as what Codex unlocks. The org-chart story is what Notion is choosing to do with that unlocked capacity, and that is the part that generalizes beyond any single tool.