At a glance
WHAT IT’S REALLY ABOUT
Hands-free AI browser control with Codex for work and life
- Codex’s desktop app plus Chrome extension enables AI to control mouse, keyboard, and web navigation via @browser, @chrome, and @computer modes.
- Claire showcases an AI-driven QA workflow that tests a web onboarding flow across desktop and mobile, captures screenshots, and outputs a prioritized Google Sheet of issues.
- She applies “persona testing” by having the agent use her product as a PM, engineer, and team lead, then produce a research-style critique highlighting friction points and missing mental models.
- Productivity examples include hands-free LinkedIn inbox triage and drafting replies despite the lack of an official LinkedIn API/MCP integration.
- Personal “digital toil” reductions include AI shopping that adds items to cart, filling tedious forms, controlling an iPhone via mirroring, and handling router/remote-access tasks while out of state.
IDEAS WORTH REMEMBERING
5 ideasTreat AI as a virtual coworker who can operate your UI.
With Codex computer/browser use, the model can click, type, resize viewports, and navigate flows end-to-end, turning manual web interactions into delegated work.
Automated QA is most valuable for edge cases you don’t naturally test.
Claire’s agent found a high-severity blocker (missing required-field validation) because it intentionally explored failure states that a human “happy path” tester might skip.
Ask for structured outputs (Sheets/Docs) to make results immediately actionable.
Having the agent compile findings into a Google Sheet with repro steps, severity, and embedded screenshots creates a ready-made bug tracker you can hand to yourself or other agents.
Persona testing surfaces expectation gaps, not just UI bugs.
Running PM→Engineer→Team lead flows revealed a conceptual break: the agent expected to reference/@mention an existing PRD across threads, exposing a product-model mismatch.
Under-prompting can improve results with frontier models.
Claire notes that broad instructions like “QA the onboarding flow” can outperform overly prescriptive checklists, because the model plans more intelligently and adapts as it explores.
WORDS WORTH SAVING
5 quotesSo imagine you have a virtual coworker, and that coworker can put its hands on your keyboard and get stuff done on your computer.
— Claire Vo
Now, this sounds really scary, but in fact, it has been one of the most life-changing features of AI.
— Claire Vo
I actually found that under-prompting these models gets you a much better effect than over-prompting them.
— Claire Vo
Sometimes we put AI to use for us, and then sometimes we need AI to put us to use, and so sometimes I am just the human verification layer on top of ChatGPT browser use.
— Claire Vo
Everything is computer, and I really think that this is gonna change how we interface with our digital lives.
— Claire Vo
High quality AI-generated summary created from speaker-labeled transcript.
