OpenAI's Codex became a general-purpose workhorse even for non-developers, which its own team didn't expect

OpenAI’s Codex was built as a developer tool, but non-developers at the company ended up preferring it over products designed specifically for their roles. Andrew Ambrosino described the outcome bluntly: “nobody would leave the Codex app for the apps that were allegedly for these other personas.” That is a notable failure mode for persona-targeted design, and one worth watching as more teams try to carve up AI products by job function.

nobody would leave the Codex app for the apps that were allegedly for these other personas Andrew Ambrosino

Part of what makes Codex sticky, according to Ambrosino, is its architecture as an orchestration layer rather than a self-contained application. It acts as a “home base” for tracking tasks across surfaces, opening other apps when needed rather than trying to absorb every workflow into a single interface. The model is closer to a coordinator than a super app.

Codex also appears to handle integration gaps on its own. When a third-party API could not give it the access it needed, the system built its own Premiere Pro extension to manipulate markers directly inside the application. That kind of self-directed tool creation is still narrow, but it illustrates how the product fills gaps that fixed integrations leave open. How broadly that capability generalizes beyond specific, well-scoped tasks remains to be seen.