What Is a Browser Bot? A Selenium Dev's Honest Switch to AI
My friend Darren spent three years writing Selenium scripts. CSS selectors, XPath queries, flaky test suites that broke every time someone moved a button.
One morning he woke up to an empty spreadsheet. A competitor redesigned their pricing page. His selector, div.pricing-card > span.amount, pointed at nothing. Bot ran fine. Just couldn’t find anything.
Two hours to fix it. For a task that takes a human 90 seconds.
That was when he started looking for something else.
Three Generations of Browser Bots
A browser bot is software that automates actions inside your browser. Darren’s gone through three generations of these tools and every one was supposed to be the last.
Scripts (Selenium, Puppeteer, Playwright) were his first real love in automation. Write code, browser obeys. He had a Puppeteer bot that scraped competitor pricing every morning, another that auto-filled vendor forms, a Playwright script that screenshot’d dashboards for standup. The power was real. But he tracked his maintenance time one quarter: 31 hours fixing broken automations. The bots were saving him maybe 20 hours of manual work in that same period, so he was underwater. Every time a site deployed a new frontend, he’d spend a weekend rewriting selectors. He got tired of it.
No-code RPA was supposed to fix that. Record clicks, replay them, no more selectors. He tried it. Felt like progress for about a week before he realized it was the same click-replay approach dressed up in a drag-and-drop UI. He also watched a company he consulted for spend $200K on UiPath licenses. Six months later half their automations were broken and the person who built them had left.
Then he tried an AI browser bot. He was on a LinkedIn profile, researching a prospect. Instead of writing a script to extract their headline and experience, he opened a side panel and typed: “Summarize this person’s background and draft a connection message.” The AI read the page. Not the selector. The actual content. It understood who the person was and wrote a message referencing their recent career move.
No selectors. No maintenance. He just told it what he wanted.
The technical term is AI browser agent. Most people still search for “browser bot” though.
The Difference That Matters
Old bots navigate the DOM. New ones actually read the page.
When a website redesigns, Selenium breaks because div.pricing-card no longer exists. An AI browser bot doesn’t care about the div. It reads the page like you would and finds the pricing wherever it ended up. Honestly it still gets confused sometimes, especially on pages with heavy JavaScript rendering or weird iframes. But it handles layout changes that would completely brick a Selenium script.
| Script Bot | AI Browser Bot | |
|---|---|---|
| Input | CSS selectors and code | Plain English |
| When layouts change | Breaks | Usually handles it, sometimes needs a nudge |
| Maintenance | Darren tracked 31 hours/quarter | Way less, but not zero |
| Who can use it | Developers | Anyone who can type a sentence |
Darren’s non-technical colleagues couldn’t touch his Selenium scripts. But they can type “fill out this form with our company info.” That’s probably the thing that surprised him most.
Where They Actually Live
Standalone AI browsers (OpenAI’s Atlas, Perplexity’s Comet) are whole new browsers with AI built in. Darren tried one for a week. The AI was good. But he kept reaching for his Chrome bookmarks, his password manager, his other extensions. Went back on day 8.
Cloud bots run headless browsers on remote servers. Good for batch scraping 500 pages overnight. Useless for daily work because they can’t see the email you’re reading right now.
Browser extensions are what Darren actually uses. AI in your existing browser’s side panel. Sees the page you’re on, acts without you leaving your workflow. The browser is where work happens, so this made the most sense to him.
Picking One
After getting burned a few times, the main thing Darren checks is whether a browser bot actually reads the page or just replays clicks. If it has a recorder, it’s the same old approach. If the answer to “what happens when the website changes” is “re-record the workflow,” that tells you everything.
Beyond that: can you choose your AI model? Being locked into one provider burned him before when pricing changed overnight. Where does your data go? A 2026 Incogni study found 52% of AI browser extensions harvest user data, which is insane. He only uses bots where he brings his own API key so data goes straight to his AI provider, not through someone else’s servers.
And can you save and reuse workflows? He doesn’t want to retype the same instructions every week.
What Actually Changed
The biggest change isn’t speed. It’s that Darren’s non-technical teammates can automate stuff now without asking him to write a script.
Selenium meant writing code, debugging selectors, handling edge cases. A dev project every time. Now someone on his team just types “fill this out” or “summarize this page” and it works. He told me he’s honestly still a little conflicted about it. He spent years getting good at browser automation scripting, and now the skill that matters is being able to describe what you want clearly. Different muscle entirely. But he can’t argue with the results. The whole team moves faster when automation isn’t bottlenecked through one developer.
That’s the shift he keeps coming back to.
If you’ve been writing scripts or doing this stuff manually, dassi is what Darren uses now. AI browser agent in your Chrome side panel. Supports 50+ models, doesn’t store your data, and you can try it free for 7 days. Get started here.