TLDRocket
Sign in

Launch HN: Intuned (YC S22) – Build and run reliable browser automations as code

intunedhq.com fkilaiwi

Intuned just launched on HN: an AI agent that writes browser automation code and fixes it when websites change. It's basically self-healing web scrapers, aimed at the endless problem of sites without APIs breaking your scripts overnight.

Every company that's ever tried to scrape a vendor portal or auto-fill a government form knows the real pain isn't writing the automation. It's keeping it alive. Selectors shift, layouts change, and suddenly a script that worked fine last Tuesday throws errors nobody can reproduce. Intuned, a startup founded by Faisal and Ahmad, is betting its whole product on solving that maintenance headache rather than the initial build.

The company actually started as something else entirely inside YC's Summer 2022 batch. Faisal's background at UiPath kept drawing questions from batchmates about whether old-school RPA techniques could patch over missing APIs, and when it came time to pivot, that question turned into the company. By early 2024 they were testing a code-first approach with a small group of customers, and it took, in their words, a fair number of iterations to land on the current shape of the product.

What they ended up building pairs a managed runtime for Playwright-based automations, written in TypeScript or Python, with an AI agent that can generate, deploy, and later repair that code. Users can write scripts themselves in Intuned's browser IDE or just hand requirements to the agent. Once deployed, each project runs in an isolated machine, and the platform handles the unglamorous plumbing: session reuse, scheduling, concurrency, batch runs, logging.

The more interesting design choice is how tightly the runtime and the agent are wired together. Every run gets logged with full context, parameters, traces, results, so when something breaks, the agent isn't guessing. It can open the failed run, inspect exactly what happened, and either propose a fix for a human to approve or push the fix automatically depending on how a project is configured. The team says they scrapped an earlier, more rigid version of this agent late last year after seeing what open-ended coding agents like Claude Code and Codex could pull off, and rebuilt around Anthropic's Claude Agent SDK instead, adding their own execution environment, a CLI, and a custom skills-and-MCP layer tuned to browser work.

They've also packaged the whole thing as an API, called Web Task API, so other products can bolt on browser automation without touching Intuned's interface directly. The pitch, in short, is that code-based automation can be as fast and cheap as everyone hopes, minus the part where it quietly rots the moment a target website ships a redesign.

My take

Self-healing automation is the right instinct, because the entire RPA industry has spent two decades pretending brittle scripts are a maintenance plan rather than a liability. Wiring an AI agent directly into the failure logs, instead of bolting AI on as a chat sidebar, is the sane version of this idea, and it's a good sign the founders scrapped their own rigid pipeline once real agent harnesses proved better. The open question is whether letting an agent auto-deploy fixes to production scrapers becomes a trust problem the moment it touches anything regulated or revenue-critical.

Read more about this at: intunedhq.com

Related stories

The daily briefing

Every AI story that matters, in your inbox by 8am.

TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.