Saffron Health just open-sourced Libretto, a toolkit that gives coding agents a live browser and a token-efficient CLI for building web automations that don't break every time someone changes a div. The tool lets agents inspect live pages, capture network traffic, record and replay user actions, and debug broken workflows interactively. It works with OpenAI, Anthropic, Gemini, and Vertex for snapshot analysis, keeping heavy visual context out of your main agent's context window. The whole thing runs on top of Playwright.
Point Libretto at a browser script and ask it to reverse-engineer the underlying API calls, converting slow, fragile UI automations into direct network requests. Demonstrate a workflow manually and the agent turns it into a reusable Playwright script. Or hand it a broken integration and let it reproduce the failure, inspect the live page, and fix the issue on its own. For Saffron Health, this came from a real pain point: healthcare software that lacks APIs entirely, forcing teams to rely on browser automations that constantly break.
The Hacker News discussion raised valid concerns about long-term reliability. The harder question is whether that script survives DOM changes, layout updates, and the general chaos of production websites over months. Community members also asked about integration with Google's Scion, hoping to avoid stacking multiple AI subscriptions. Libretto delegates snapshot analysis to a separate model call, rather than burning tokens in your main coding agent. That's a smart architectural choice that directly addresses cost concerns.