At a glance
WHAT IT’S REALLY ABOUT
GPT-6 Astra excels at computer use, QA, and ambitious builds
- Claire Vo reviews early access to GPT-6 “Astra,” calling it a step-change model that one-shots tasks prior models failed and becomes her new daily driver.
- The standout capability is computer/browser use: Astra can navigate complex UIs (CRMs, node-based tools, Figma) to build workflows, generate assets, and assemble deliverables with minimal human input.
- She shows Astra acting like a persistent QA assistant by testing a preview web app, inspecting console logs, and repeatedly triggering edge cases such as race conditions over a long run.
- On coding, Astra allegedly cracked a hard ChatPRD “product intelligence” feature involving multi-source ingestion, deduped insights, and an auto-generated product wiki verified against external sources.
- She demonstrates “ambition unlocks” beyond web apps: hardware hacking via a Divoom MiniToo CLI/live display, a Mac AIM-style Codex client, and Blender-driven 3D prototypes including her ‘Barbie Bench’.
IDEAS WORTH REMEMBERING
5 ideasAstra makes UI-based automation practical again (even in complex, node-heavy tools).
Claire’s strongest claim is that Astra reliably navigates complex, real-world web apps (node builders, many fields, messy layouts) without constant user babysitting, turning “I should automate this” from a weekend project into a quick delegation.
Browser/computer use can implement business-process workflows end-to-end, not just click around.
She demonstrates Astra modifying a CRM workflow (Adio) to route leads and generate personalized outbound email drafts authored as either Claire or her partner, then posting the draft into Slack—work she previously spent an hour manually fiddling with.
Creative pipelines shift from manual production to human taste checks and final polish.
In Flora, Astra imports new photos, selects the image model (gpt-image-2), applies existing prompting patterns, and generates stylized thumbnail-ready assets; she frames the human role as taste/QA rather than repetitive node-dragging.
Astra is a strong “QA agent” that can execute long, systematic browser testing with console inspection.
Claire uses computer use for software QA: Astra opens a preview branch, exercises flows, inspects console errors, refreshes to trigger race conditions, and runs extended sessions (~1h45m) to uncover edge cases—tedious but valuable human work.
For complex engineering tasks, Astra can jump from ‘architectural flailing’ to shippable quality in fewer prompts.
Her flagship coding example is a “product intelligence” feature for ChatPRD: ingesting multi-source unstructured data, deciding durable storage, deduplicating insights, generating an auto-wiki, and verifying against external sources—work she says other models couldn’t reach high quality on, but Astra reached ~90% in one pass.
WORDS WORTH SAVING
5 quotesThis has been the one that has made me actually feel more ambitious.
— Claire Vo
UI's back, baby. Like, SaaS, this might be your savior because humans really do like a button, and now that you can get AI to click buttons for you and use a UI for you, you sort of don't have to think about these CLIs and MCPs necessarily if the right expression of your product is more of a UI-based thing.
— Claire Vo
I do spend about 90% of my day sitting here watching Astra use my computer and specifically use Chrome to navigate websites and applications that I really love.
— Claire Vo
I am telling you all, Astra one-shotted.
— Claire Vo
This is my Everest.
— Claire Vo
High quality AI-generated summary created from speaker-labeled transcript.
