Build it

Automating the browser: scripts, extensions, and how to do it ethically

July 29, 2026 · 8 min read

An enormous amount of business work happens inside a browser tab: checking competitors' public catalogs, gathering supplier information, filling web portals. All of it can be automated with JavaScript and browser extensions — code an AI will happily write for you. Here's what that unlocks, and the ethics that separate a professional from a nuisance.

What browser automation actually is

Websites are structured documents. A short piece of JavaScript can read a page the way you do — find the product names, the prices, the availability — and collect them into a spreadsheet instead of your short-term memory. Package that code as a small browser extension and it becomes a button: navigate to a page, click, data lands in a file.

Until recently, writing that code required a developer. Now you describe what you want to an AI — "on pages like this one, collect the part number, description, and price into a CSV" — and it writes the extension. You're the director; the AI is the builder, same as with Python.

What it's good for in a real business

• Competitive awareness. What do competitors publicly list, at what price, in what categories? That's public information you could read by hand — automation just makes the reading systematic instead of anecdotal.
• Catalog gap analysis. Compare what a prospect carries against what's available in the market and you have an evidence-based sales conversation. In my own prospecting, this exact pattern — automated gap-finding feeding auto-generated presentation decks — turned a day of prep into minutes.
• Portal chores. The same web forms, statuses, and downloads you check every day can check themselves and report back.
• Public records and reference data. Government databases, spec sheets, published directories — public data that's useful in aggregate but painful to collect by hand. My NHTSA project is this idea at scale.

The ethics section (not optional)

This is where I'm going to be more direct than most articles on the topic, because doing this wrong is both easy and consequential. The rules I follow, and teach:

• Public means public. Automate only what you could legitimately read by hand in a normal browser: public catalogs, published prices, open directories. If it's behind a login you don't own, a paywall, or any access control — it is not "publicly available," and automating past that line is where scraping stops being research.
• Respect robots.txt and terms of service. Most sites publish rules about automated access (a file called robots.txt, plus their terms). Read them. If a site says don't, the professional answer is: don't. There's usually another source for the same data — often an official one.
• Rate-limit like a guest. A polite scraper behaves like a fast human, not a flood: one page at a time, pauses between requests, running off-peak. Hammering a small business's website with hundreds of requests a minute can genuinely degrade their service — that's not clever automation, that's being a bad neighbor with extra steps.
• Collect facts, not people. Product data, prices, specs: fine. Harvesting personal information is a different activity with real legal exposure (and it's also just not who you want to be). If a workflow starts pulling names, emails, or profiles in bulk, stop and get proper advice.
• Prefer the official door. Many sites offer data feeds, exports, or APIs precisely so people don't scrape. The official route is more stable and more defensible — check for it first.

The test is simple: if the site's owner watched your automation work, would you be comfortable explaining it? If yes, you're doing research. If you'd reach for excuses, you already know the answer.

Where AI fits in the pipeline

Two places, and it's worth keeping them straight. First, AI writes the collection code — one-time help, no ongoing cost. Second, and only when needed, AI interprets what was collected — because scraped pages are often messy, and "which of these three hundred product listings are actually comparable to ours?" is a judgment call. That's a classic hybrid workflow: deterministic collection, AI interpretation at the single messy joint, human decision at the end.

A note on fragility

Honest limitation: websites change, and when they do, your script breaks. This used to mean calling a developer. Now it means a five-minute conversation with the AI that wrote the code: "the site changed, here's what the page looks like now, fix the extension." Maintenance hasn't disappeared, but it's shrunk to something a non-programmer genuinely handles.