A website with no API,
operated carefully, inside a tested script
Plenty of necessary work still lives on a website with no API. Checking a competitor's price. Downloading a monthly report. Submitting a listing on a portal that only offers a web form. This agent operates that site the way a person would, clicking and typing inside a tested, rate-limited script. It stops, rather than bypassing a CAPTCHA or a login wall it was not built to get past.
The repetitive work no API covers
Some of the most tedious recurring work in a business happens on a website with no API at all. A government portal. A legacy vendor dashboard. A competitor’s public pricing page. The task is simple in principle: click here, read this, type that there. Doing it every day or every week by hand is exactly the kind of repetitive work that nobody enjoys and everybody eventually stops doing consistently.
The second cost shows up when this work gets scripted informally. It often gets done carelessly, hammering the target site with requests. That risks getting blocked, or worse, causing real problems for a site that was not built to handle automated traffic at that rate.
Fragility is the third. A site’s layout changes, a button moves, and a script built without any error detection keeps running anyway. It clicks the wrong thing or silently fails, which is worse than not automating the task at all.
What the agent does on the page, and where it stops
The agent operates the target site the way a person would: reading what is on the page, clicking, typing, navigating. It works inside a script that is built and tested against the real site before it runs unattended. It respects rate limits that make sense for that specific site, no rapid-fire requests. It records every run, so there is a full trail of exactly what it did.
When the site’s layout does not match what the script expects, it stops rather than guessing, and flags the run for a human to check. This matters more than speed. An automation that fails loudly and safely is far more useful than one that fails silently and wrong.
Typical scope: form submissions, price or availability checks, report downloads, portal data entry, wherever a task is well-defined but the site offers no API. Anything that would require bypassing a CAPTCHA or bot detection is explicitly out of scope.
What a human still logs in for
Logging in where a login is genuinely required needs a human in the loop. So does any step involving a payment or an irreversible submission, unless your team has explicitly approved otherwise. Deciding the task is worth automating in the first place, and reviewing the target site’s terms of service, is a step we do together before building anything.
How the script stays within bounds
Rate limits are set to respect the target site, never to hammer it. The agent never bypasses a CAPTCHA or bot detection. It stops and hands off instead. Every run is recorded, screen and action log, so it is fully auditable. A layout mismatch halts the run rather than letting the agent proceed on a guess.
Price and timeline
| Option | Price | What it covers | Timeline |
|---|---|---|---|
| Agency runs it | from $1,800 + support plan | Script built, tested and maintained by us, monthly run review | 2 to 3 weeks |
| Full control, handover-ready | from $2,800 | Same script on your own infrastructure and credentials, documented, your team runs it | 3 to 4 weeks |
Running cost is usually $15 to $60 a month in model and browser session usage, depending on run frequency.
Related
See the AI agents service page and automation-everything for the surrounding build. Within this group, computer-use agent for legacy desktops applies the same screen-level approach to desktop software. Scraping and research agent and API integration agent cover adjacent ground. For a related one-time setup, see automate a computer-use agent for legacy systems and automate price monitoring. Real governed-agent discipline behind this page: the ProBay AI agent team case study.
A task stuck on a site with no API? Get in touch and we will check if it is a good fit for this first.
FAQ
How much does a browser automation agent cost?
From $1,800 for one target site and a defined task, live in 2 to 3 weeks. Multiple sites or more complex, multi-step tasks usually run $2,800 to $4,500.
How long before it is running the real task?
2 to 3 weeks. Scripting and testing against the actual site takes most of it. A site's layout can shift in small ways that need handling before the agent runs unattended.
Which sites can it work with?
Most sites a person can log into and operate with a browser: portals, dashboards without an API, government or industry sites, competitor pages for public price checks. We check the site's terms of service before committing to a build.
What happens if the site changes its layout or something goes wrong mid-task?
It detects the mismatch and stops rather than guessing and clicking the wrong thing. The run is flagged for a human to check, and the script gets updated once the new layout is confirmed.
Does this bypass logins, CAPTCHAs or bot detection?
No, never. It operates with credentials you provide, the same way a person would. It stops at a CAPTCHA or bot-detection wall rather than trying to get past it. Where a login is required, a human handles that step.