Notion, the company known for its notes and documentation workspace, has put OpenAI's AI coding tool Codex at the center of its development work, reshaping how its engineers operate[1]. In one case, work that would once have taken about two weeks was completed in roughly three hours[1]. The account offers a concrete look at a workplace where code written by AI agents is becoming the norm.
A Two-Week Feature Built Solo in Hours: Shipping Voice Input on the Web
Ryan Nystrom, who runs AI Product Engineering at Notion, says his team has touched nearly every AI feature in the product over a tenure of just over a year[1]. He describes how he brought voice input to Notion's web client using Codex exclusively[1].
When the request came in, the mobile app already had a working version, but the desktop and web clients did not[1]. Nystrom himself was not entirely sure how the feature worked on mobile, yet he pointed Codex at the mobile codebase, gave it a clear description of how it should look on the web, and provided a way to verify the result[1]. Codex then returned a complete first cut of the web implementation in a single shot, and because it matched Notion's coding conventions, the result was ready to ship the next day[1].
"If I were to build the Notion voice input feature two years ago, this is a project that would've taken me and maybe another engineer two weeks," Nystrom says, adding, "With Codex, I was able to build this in maybe three or four hours, entirely by myself"[1]. He values how Codex first explored the mobile code carefully before writing nearly the entire feature in one shot: "What I appreciate about Codex is that it takes its time to figure things out before actually building. The result is that usually what it builds is to our codebase's standards off the bat, rather than me having to go back and clean up a bunch of its work"[1].
"I Don't Really Write Code by Hand Anymore": How the Work Has Changed
Notion's engineers can hand Codex a set of tasks along with a way to check the results, then step away to other work[1]. "I've almost found myself spending a lot more time writing these spec documents that I can hand to Codex and let it work on," Nystrom says. "Honestly, I don't really write code by hand anymore"[1].
The change in workflow is concrete. Before Codex, each engineer could realistically focus on only one task at a time, squeezed between meetings and supporting peers[1]. Now they run multiple tasks in parallel while keeping the team support that used to be the bottleneck[1].
Codex also works around the clock. Nystrom describes posing a research question before bed, letting Codex run overnight, and waking up to a finished report[1]. "Whenever I need to research a task, fix a bug, or make a little tweak, Codex is just there, ready and willing. Basically, I've got an intern available at Notion 24/7," he says[1].
Managers Back in the Codebase, and a New Baseline for Small Teams
The shift is not limited to engineers. "I manage a team of people, and traditionally managers haven't had time to write code," Nystrom says, adding, "The fact that I can build a feature solo while still supporting my team is crazy. I've been managing for five-plus years and never been able to go this deep on coding problems"[1]. He can now queue up a task, head into a block of meetings, and return to a finished feature[1].
OpenAI frames Codex as having "reset the baseline for what a small team (even a team of one!) can ship"[1]. At Notion, the company is rethinking the software primitives and abstractions it builds so that agents can use them more easily[1]. When hiring new engineers, it is looking for curiosity and open-mindedness, since the years of experience the field would normally call for do not yet exist[1].
Summary
Notion's use of Codex offers a concrete picture of how AI coding agents are changing the assumptions of software development. A two-week feature is built solo in a few hours, managers are returning to the codebase, and a way of working in which engineers write specs and hand them to an agent is spreading. As time spent writing code by hand declines, the weight shifts toward designing what to build and verifying the results, and that is the heart of the change described here.
