Codex OS automates 80 to 90 percent
Head of Product at SaaS startup on building a personal AI OS with Codex automations and Claude Cowork
This is less about one app replacing every tool, and more about work becoming recordable, searchable, and re-routable through an agent layer. The jump to 80% to 90% happens because meetings, calls, inboxes, Slack threads, bugs, and calendars are all connected back into Codex, so it can surface commitments, draft follow ups, reschedule work, and run recurring scans. The main holdout is original writing, where the operator still wants the thinking and final voice to stay human.
-
The real behavior change is total capture. Call recordings, transcripts, messages, emails, and calendar activity all feed the system, including live conversations after the fact. That gives Codex enough raw material to build workflows like what I promised yesterday, which turns spoken commitments into an action queue.
-
Coverage does not mean every click happens inside Codex. Slack is still often answered in Slack, and the calendar UI is still used directly, but Codex becomes the layer that reads across tools, drafts work, prioritizes the day, and triggers actions through connected accounts and browser automation.
-
This pattern depends on product maturity and custom plumbing. The operator connected four email accounts, four Slack accounts, multiple calendars, iMessage, Grain, Linear, Figma, Chrome, and Google Workspace APIs, and still spends about 14% of usage maintaining automations. Claude stays in the loop mainly for higher level reasoning through non interactive handoffs.
The next step is not broader access, it is better memory and better voice control. As agents get better at remembering constraints, learning from edits, and staying in character on messages, the remaining manual layer shifts from drafting and routing into only the most personal writing and the highest judgment decisions.