Repository navigation
one form per spreadsheet row
Take a spreadsheet, submit one web form per row. It is the oldest office automation task there is, and it is the one where an agent's mistakes are most expensive, because the run repeats them at machine speed with nobody reading.
Three things decide whether this works: what it costs, how you recover, and how you stop a bad run at row one instead of row two hundred.
Typing is the dominant cost and it is not close. Measured on this browser, a field costs roughly one second plus 270 ms per character, with the full numbers on how long an AI browser agent takes per step.
So a form with a name, an email, a company and a short message, say 120 characters across four fields, is about 36 seconds of typing plus four seconds of field overhead plus a second for the submit. Call it 45 seconds a row before page loads and before the model thinks.
Two hundred rows is two and a half hours. That is a fine answer if you expected it and an unpleasant surprise if you budgeted by the number of rows rather than by the number of characters. Count characters, not fields.
The reason typing is paced rather than instant is in that page too: a field that fills instantly with no inter-key variation is one of the cheapest things for a page to notice, so the cost buys the thing you came for.
Add a column to the spreadsheet, and have it updated after each successful submission with the timestamp and whatever the page gave back as a confirmation.
The updating is done by your client rather than by the browser: this tool surface reads and acts on pages and has no save-file tool, so the file work belongs to whatever you are driving from. That division, and the ceilings it imposes on a long run, are in extracting data to a CSV with an AI agent.
That single column gives you:
- Resume. A run that dies at row 83 restarts by skipping rows that have a value.
- Proof. A reference number per row, so "did this one go through" is a lookup rather than an argument.
- Duplicate protection. The instruction "skip any row that already has a confirmation" is what stops a rerun from submitting everything twice, and a double submission is the most common real-world damage from this task.
Without it, a failed run leaves you with a spreadsheet and no idea which half of it is already in the other system. That is worse than not having run it.
Before releasing the list, submit exactly one row and go and look at the result in the destination system. Not the agent's report of the result: the record itself.
This catches the class of error that is invisible from the browser side and fatal at scale: a field that silently truncated, a date the form parsed as month-first, a dropdown that defaulted because the value did not match an option, a required field that was optional-looking and got left blank.
Every one of those produces a successful submission and a wrong record. The agent has no way to know, because the page said thank you.
Dropdowns. A select needs the option's value, not the label you have in the
spreadsheet. "United Kingdom" in your column may be GB in the form. Decide the
mapping before the run, and have the agent record the option it actually chose,
not the one it intended.
Dates. The same eight characters mean two different days depending on the form's locale. Write the format you expect into the task and have the agent read back the field after filling it.
Fields that reformat as you type. Phone numbers and card-shaped inputs insert their own separators, so what you typed and what is in the field differ. Read back rather than assume.
Fields that appear conditionally. Choosing a country reveals a state dropdown. A task written against one row's shape breaks on a row with a different country, and the failure looks like a missing element.
The general principle behind all four: read the field back after filling it, before submitting. It is a free read, and it converts four silent corruptions into four visible errors.
Open the form at the URL below once and keep the session. For each row of rows.csv where the "submitted" column is empty:
Fill the fields from the row. After filling each one, read it back and check it matches what you intended. If any field does not match, write the difference into the "submitted" column, prefixed with "MISMATCH:", and move to the next row without submitting.
Otherwise submit, wait for the confirmation to appear, and write the date and the confirmation text into the "submitted" column. Never submit a row that already has a value in that column.
Stop and tell me if three consecutive rows fail.
A form that submits two hundred times in two hours sends two hundred notifications to whoever is on the other end. If that is your own system, fine. If it is somebody else's, the volume is the thing they will notice and the thing they will mind, and the mechanism is the one in agent retry loops trip rate limits, not fingerprints.
Pace it, and if the task is contacting people rather than filing records, the question of whether to send two hundred of anything is a separate one that a browser does not answer.
How long does it take to fill a form with an AI agent? About a second per field plus 270 ms per character, plus page load and model turns. Budget by total characters, not by row count.
How do I stop it submitting twice? A column the agent writes a confirmation into, and an instruction never to touch a row that already has one. Nothing else survives a rerun.
Why is the data wrong in the destination but the agent said success? A silently truncated field, a mismatched dropdown, or a date parsed in the other order. Read every field back before submitting and run the first row alone.
Can I skip the read-back to go faster? The read costs a hundredth of a second. It is not where the time is.
The form changes depending on what I pick. Expect it: conditional fields are normal. Write the task against the branches you have in the data, not against one example row.
See also: getting an AI agent to fill out forms for the single-form version and the mechanics of each control type, and keeping an AI browser agent out of destructive actions, because submit is irreversible and this task does it in a loop.
- The typing cost model, about one second fixed plus 270 ms per character, is measured and sourced on how long an AI browser agent takes per step: 1.87 s for 3 characters, 3.93 s for 12, 12.63 s for 43.
- The field-level failure modes are from running forms with this browser and are offered as experience.
Run row one alone and go and look at the record. It is thirty seconds and it is the only step here that catches the mistakes the browser cannot see.
- OpenAI Operator alternatives
- Open-source Operator-style agents
- Is OpenAI Operator still available?
- OpenAI Operator vs Claude computer use
- browser-use alternatives
- Choosing an AI browser agent
- Open-source AI browser agents
- Open-source computer-use agents
- What is an AI web agent?
- AI browser agents vs traditional scraping
- Cloud browser infrastructure for AI agents, explained
- Browserbase alternatives
- Firecrawl vs an AI browser agent
- Skyvern alternatives
- Stagehand vs browser-use
- Project Mariner is gone: what replaced it
- Manus alternatives
- Gemini computer use vs Claude computer use
- A stealth browser MCP, reviewed honestly by its own wiki
- AI browser vs AI browser agent: which one do you want?
- AI browser agent vs RPA: which one fits the job
- AI browser agent vs n8n, Zapier and Make
- Vercel agent-browser alternatives, compared honestly
- What is an agentic browser? Definition and the two kinds
- Open-source agentic browsers: the three layers, compared
- Choosing an MCP server for browser automation: four axes
- Stealth MCP servers compared: Camoufox, nodriver, Patchright
- Playwright MCP alternatives, and the three you don't need
- Autonomous browser agents: the four rungs of autonomy
- What is actually free in the AI browser agent stack
- browser-use on GitHub: what the repo actually gives you
- Playwright MCP vs Chrome DevTools MCP: different jobs
- How to choose among MCP servers: a map by category
- Which MCP servers are worth adding to Claude Code
- MCP on GitHub: finding servers and judging them fast
- MCP vs an API: the decision, and what the wrapper costs
- MCP alternatives: when the protocol is the wrong shape
- Why does my AI agent get blocked?
- The timing signal AI agents give off
- Agent retry loops trip rate limits, not fingerprints
- Claude computer use detected as a bot
- browser-use getting blocked: what you can and cannot change
- Playwright MCP session blocked: four causes, four fixes
- Playwright MCP and captchas: what actually gets you past
- Cloudflare and a browser MCP server: what is being read
- Can an AI agent solve a captcha? The honest answer
- Getting an AI agent to fill out forms
- The best model for an MCP browser agent, and what it really costs
- Browser problem or model problem?
- How to use a Playwright MCP server with Claude Code
- Extracting data to a CSV with an AI agent
- Monitoring a page for changes with an AI agent
- How to let Claude Desktop control a browser
- How to add a browser to Cursor as an MCP server
- Using an AI agent to hunt for apartments
- Getting website data into Google Sheets with an AI agent
- Using an AI agent to download invoices from portals
- AI agents for web research
- Using an AI agent to test your own website
- How to add a browser to Cline as an MCP server
- Posting to social media with an AI agent
- Posting to Facebook with an AI agent
- Posting to Instagram with an AI agent
- Posting to X with an AI agent
- Automating LinkedIn posts: read this first
- Appointment bots: what they are and what an agent can legitimately do
- Track prices across sites with an AI agent
- Build a lead list with an AI browser agent
- Run an AI browser agent on a schedule
- AI browser agent with a local LLM: what changes
- Should you log your AI agent into your accounts?
- How to write a task an AI browser agent can follow
- Move data between two web apps with an AI agent
- The MCP server
- How the tools are shaped, and why
- Playwright MCP vs the Playwright CLI: which fits when
- Playwright MCP: browser is already in use, and the fix
- Playwright MCP best practices: four decisions that matter
- Playwright MCP with a proxy, and the three leaks it leaves
- A browser MCP server in GitHub Copilot: setup and limits
- Using a browser MCP server for web scraping: the pattern
- Which LLM for browser automation: the four properties
- How to build a browser agent, and what to take instead
- Getting an AI agent to log into a website: three routes
- MCP tools, resources and prompts: who controls each
- How many MCP tools is too many? The context arithmetic
- How to build an MCP server: the decisions, not the scaffold
- Local or remote MCP server: what changes, and what does not
- Writing an MCP client in Python: the thirty-line version
- Self-hosted AI agent: what one actually costs to run
- How long an AI browser agent takes per step, measured
- Text, HTML, snapshot or screenshot: what the agent should read
- Giving an AI browser agent a stopping condition
- Keeping an AI browser agent out of destructive actions
- Why did the AI agent click the wrong thing
- When the page changes under the AI agent
- Running one AI agent task across a list of sites
- Seeing a page as it appears in another country
- Getting data out of a dashboard with no export button
- Two browsers in one session: main and support
- Finding the dead links on a site with an AI agent
- Filling a CRM record from a company's website
- One form submission per spreadsheet row, with an AI agent
- Dated screenshots of a page as evidence
- Checking order and delivery status with an AI agent
- Reading a PDF that opens inside the browser
- Summarising a long page or thread with an AI agent
- Collecting every image on a page with its caption
- Collecting event and course listings with an AI agent
- Cancelling a subscription with an AI agent
- What an AI agent can and cannot do inside an iframe
- Shadow DOM and an AI agent: you can click it, you cannot read it
- How to add a browser to Codex as an MCP server
- What a page snapshot costs, per control
- How to let Gemini CLI use a browser
- Native selects and the ones that only look like selects
- Clicking by selector or by coordinates
- How long the agent waits before it gives up
- What a second browser costs
- Uploading a file with an AI agent, and why this one cannot
- Watching the agent work, and when it is worth it
- When not to use an AI browser agent
- Agent or script: deciding once instead of every time
- Using the keyboard instead of the mouse
- Secrets in an agent task: where they end up
- What an agent run should log
- Deduplicating what an AI agent collects
- Normalising values across sites
- Validating an AI agent's output
- Reading a table with an AI agent
- Driving a site's own search and filters
- The task works headed and fails headless
- How to let the AdaL CLI use a browser