The conventional read on the April 16 Codex release is that OpenAI added computer use, an in-app browser, and a bunch of plugins. That’s the feature list. The feature list isn’t the story.
The story is that OpenAI now has a credible path into every piece of software with a graphical interface, and they got there without needing a single vendor to cooperate. The agent just drives the interface you already drive.
Think about what that changes. Six months ago, any software that didn’t have an API was outside the automation conversation entirely. Internal dashboards, legacy vendor portals, the app your company built in 2019 and never maintained. That software just came back inside the conversation, through a door that doesn’t require anyone’s permission to open.
Anthropic made the same bet on getting agents out of the coding box, but through structured interfaces that depend on the ecosystem building for agents first. Both labs are headed to the same destination. The paths they chose are fundamentally different, and the difference determines which software your agents can actually reach today.
Here’s what’s inside:
The mechanism nobody’s framing right. Why computer use changes what’s automatable and why structured integrations carry a dependency most people are underweighting.
Who actually built this. The twelve-person team whose line from Apple Shortcuts to Codex explains why this works when prior computer-use demos didn’t.
Two theories of where agents are going. Chronicle vs. Conway, ambient context vs. event-driven agents, and what each tells you about the next eighteen months.
When to use which. The specific capability where the gap is wide enough to change your stack, and the capabilities where Claude still wins.
What to watch. Two signals that will tell you which path is paying off before consensus catches up.
Three prompts to act on this. A workflow audit that maps where to deploy each agent, a dependency assessment for your automation stack, and an acquisition signal tracker you’ll reuse every time a deal drops.
I’ve been testing both side by side for weeks. The gap on computer use is bigger than I expected.
