Notion has been using Codex to speed development and change how engineering work is organized, according to Ryan Nystrom, who leads AI Product Engineering at Notion. He says the tool often explores existing code, produces implementations that match the codebase’s conventions, and can run tasks autonomously when given a clear specification and a way to verify results.
Porting voice input to the web
The team used Codex to bring Notion AI voice input from mobile to the web. Nystrom described giving Codex access to the mobile codebase, a description of the desired web behavior, and verification steps. Codex then returned a complete first cut of the web implementation that aligned closely with Notion’s standards, allowing the feature to ship the next day.
According to Nystrom, a project that would have taken two engineers about two weeks before Codex was available took him alone roughly three to four hours with Codex. He said the model spent time exploring the mobile implementation before producing the web version in one pass.
Changes to team workflow
Nystrom reports that engineers now frequently hand Codex a set of tasks and verification criteria, then let it run while they attend meetings or work on other items. This has allowed individuals to queue multiple tasks in parallel rather than focus on a single task between interruptions.
Managers who had stopped writing production code are able to build and ship features again, Nystrom said, because Codex reduces the hands-on time required. He also described using Codex for overnight research runs and waking to finished reports, and compared the tool to having an always-available junior engineer to handle routine tasks.
Notion’s account attributes faster delivery and increased parallelization of work to Codex when supplied with clear specifications and checks; the company reports concrete outcomes such as the reduced timeline for the voice-input web feature and next-day shipping after Codex produced a production-ready first draft.
Original source: OpenAI News