Most browser automation tools are session-oriented. You spin up a browser instance, your agent talks to it, and concurrency means managing multiple sessions somewhere. That coordination complexity has to live somewhere and it usually lands in the wrong place.

We do use a daemon, but the daemon's job is different from what you might expect. It owns the state and connection management so the agent doesn't have to. The result is that from the agent's perspective there is no "current tab". Every tab is the current tab. You just operate on whichever tab you want with the tab id, concurrently, without any session bookkeeping.

This felt like the right model for agents specifically. Current tab is a human concept. Humans need it because we have one focus at a time. Agents don't. The stateless design is what makes this possible without the daemon becoming a coordination bottleneck.

Comments URL: https://news.ycombinator.com/item?id=46971995

Points: 2

# Comments: 0

As the title says, I built this because AI-generated web automation scripts are often full of hallucinated CSS selectors. They’re simply guessing because they can't see the actual page.

My tool, Slash, fixes this by giving the agent a real browser.

It sees the page: Before writing any code, it loads the site and analyzes the live DOM. This leads to much more reliable Playwright scripts.

You get the code: It outputs the full script, not just a data file. This means no black boxes. You can run it, put it in a cron job, and modify it however you want.

The goal is to get the benefits of AI for automation without sacrificing control and verifiability.

Since we're here, I made a short demo of Slash scraping the HN front page: https://slash.cool/slash-demo-hn.mp4

I'm here to answer questions and would love your feedback.

You can try the tool itself here: https://slash.cool/

Comments URL: https://news.ycombinator.com/item?id=45135315

Points: 1

# Comments: 2

No user is going to sit and watch the agent working for that long. They'll switch to other tabs or just leave. When the task is finally done, it's showing the results to an empty browser tab.

I wanted a simple way to bring them back, without the complexity of a full-blown notification system. So I built this: an API to send a "just right" notification when a task is complete. It can be a browser sound for shorter waits, or a browser push/email for the really long ones.

It's a side project born from this pain point. The goal is to be the simplest possible tool for this one job.

The website: https://notifyon.app

I'd love to get your feedback.

Comments URL: https://news.ycombinator.com/item?id=45111794

Points: 1

# Comments: 0

The market didn't suddenly get flooded with juniors because we promoted coding. It was because the demand was astronomical. Companies were desperate for people, so bootcamps emerged to fill the gap. Why would you blame the supply for a problem created by overwhelming demand?

First, not every good idea needs to be a "secret that could change the world." The most valuable ideas often aren't glamorous; they're the ones that solve real, boring problems. If everyone only chased the next big thing, we'd never get the crucial, unsexy stuff built. For example, a new ad-tech algorithm might feel exciting, but a simple internal tool that cuts a developer's daily grunt work by an hour can have a far bigger, more tangible impact. There's value in both.

Second, the best ideas rarely come from isolated genius. You can get stuck in your own head, convinced you've found something new, only to later realize you missed a crucial piece of context. The real magic happens through constant communication. For instance, I once spent weeks on a technical solution for a problem, only to show it to a product manager who, in two minutes, pointed out a much simpler, non-technical way to solve the same user pain (he just told me to skip it). Talking with people, especially those outside your niche, helps you get a reality check, refine your thoughts, and avoid building something no one actually needs. My best ideas weren't born in a flash of solo inspiration, but in messy, back-and-forth conversations.