Intuned, a YC S22 startup, launched on Hacker News with an "Intuned Agent" that turns a plain-English task into production-ready Playwright code, deploys it, and repairs it when the target site breaks.
The interesting design choice is what it does not do. Rather than run a vision model through the browser on every execution, Intuned generates deterministic code once and only reaches for AI when a run fails, reading the trace and writing a fix. Its Web Tasks API makes the trade-off legible: a sample run scraping the YC directory returns 208 companies in 58 seconds for about twelve cents of AI plus a sliver of compute. A reuse key lets the same task get cheaper each run as it caches the learned path.
It is a pointed counter to the "agent browses live" orthodoxy of Computer Use and its cousins, which Intuned still supports as an option. The bet is that for scraping and RPA at scale, predictable code plus occasional self-healing beats paying a model to re-derive the page every time.