Key Takeaways

  • OpenAI's Andrew Ambrosino aims to make Codex the “best desktop app that has ever existed,” serving as a central hub for all knowledge work, not just coding.
  • Its internal adoption is striking: nearly 100% of OpenAI employees across all departments use Codex weekly, proving its broad utility beyond engineering tasks.
  • Codex's unique approach to integration involves AI building its own extensions for specialized external applications like Adobe Premiere Pro, rather than forcing all work into one app.
  • The vision is for Codex to be where you “start work, you end work, you automate work,” intelligently using any tool needed to complete a task.

The 'Home Base' That Everyone Actually Uses

Andrew Ambrosino, product and engineering lead for OpenAI's Codex app, doesn't think small. His goal? To make Codex the "best desktop app that has ever existed." Forget niche developer tools; Ambrosino describes a future where Codex becomes the central "home base" for all knowledge work.

This isn't just aspirational marketing. Internally at OpenAI, Codex already sees nearly 100% weekly usage across every department. That's not just engineers. As Ambrosino notes, “Internally at OpenAI, nearly 100% of their employees use Codeex weekly. And that is not just the engineers.” This widespread adoption proves its value extends far beyond code generation, acting as a general-purpose assistant. It's designed to be where “you start work, you end work, you automate work, and it uses whatever you need to do,” as Ambrosino puts it. This vision challenges the conventional wisdom that specialized tools should live in silos.

AI That Builds Its Own Integrations

The real insight isn't just that Codex helps with diverse tasks, but how it does it. Instead of forcing users to stay within the app, or building every feature itself, Codex operates as an intelligent orchestrator. It understands when to call upon external, specialized tools. But here’s the kicker: it can even create those bridges itself.

Ambrosino shared a particularly sharp anecdote. An OpenAI employee wanted to know if Codex could edit videos. Codex isn't a video editor, of course, but it understood the user's intent. It knew the user worked with Adobe Premiere Pro. Initially, Codex tried to edit the Premiere Pro project files directly. That worked for some tasks, but not all. Recognizing its limits, Codex then did something unexpected: it built itself a custom extension for Premiere Pro. This extension could be installed into the video editing software, allowing Codex to then "talk to" Premiere Pro and say, "Hey, Premiere Pro extension, can you please change this marker inside of the Premiere Pro app."

This isn't simply calling an API. This is an AI agent assessing a capability gap, understanding the need for deeper interaction, and then generating the solution (a custom extension) to bridge that gap. It signals a shift from rigid API integrations to AI agents dynamically building the tools they need to complete complex tasks, using external applications as extensions of their own capability.

What to Do With This

Stop thinking about AI as a feature or a single application. Instead, picture it as a super-intelligent coordinator, the "home base" for your entire operation. This week, pick one multi-app workflow that eats up too much of your team’s time – say, taking data from a CRM, running it through a spreadsheet, then drafting a personalized email. Instead of trying to find one tool that does it all, ask how an AI agent, like Codex, could orchestrate this. Could it identify needed external tools? Could it build helper scripts or even temporary browser extensions to automate the clicks and data transfers required, even without native APIs? Begin to design your internal tools and processes with this adaptive, bridge-building AI in mind, rather than waiting for perfect, pre-built integrations.