I was scrolling Indeed last week looking for something else, and the third result was a “Data Entry Specialist” role at $19/hr that listed, as a primary responsibility, “transferring information from PDF invoices into NetSuite.” The posting went up in April 2026. Someone, somewhere, is reading this paragraph during their lunch break right now, having just finished invoice 47 of 200.

Manual data entry was supposed to be dead by now. RPA was going to handle it. So was OCR. Then APIs, then no-code tools, then ChatGPT. And yet the Bureau of Labor Statistics still tracks “data entry keyers” as an occupation, with around 175,000 of them in the US — plus an enormous shadow workforce of admins, ops people, and analysts whose actual day is mostly typing what one screen says into another screen.

The math is grim

A salesperson I know spends roughly 90 minutes a day moving meeting notes from Gong into HubSpot. That’s 7.5 hours a week. About 32 hours a month. Almost a full workweek every month, gone on a task that is, mechanically, indistinguishable from what a temp would have done in 1998.

She knows it’s stupid. Her manager knows it’s stupid. But the HubSpot integration with Gong exists in some half-baked form that doesn’t capture the custom fields her team needs, so somebody has to manually copy the call summary, paste it into the Notes field, find the deal stage from the recording, click the right dropdown, and save. Multiply that by every salesperson at every CRM-using B2B company on earth.

Why nobody fixed this

Because the work sits in the gap between systems that don’t talk to each other, and the gap is too narrow and too specific for any single vendor to bother bridging it.

API integrations work when both sides cooperate. But they break the moment one side adds a required custom field, a permissions wall, or a tab that isn’t reachable without clicking through three modals. Every ops team I’ve talked to has a Notion doc full of exceptions explaining why the “automated” workflow still needs a human babysitting it.

So companies tried RPA. Record the clicks, replay them, scale up. And the tools turned out brittle as hell: change a button color, kill the workflow. Maintenance ended up costing more than the labor it replaced.

What works

Show me the two tabs. Tell me what to move. The whole loop is read here, type there, flag the edge cases.

A different shape of fix

A browser agent that can read the page you’re on, type into a different page, and respect your already-logged-in sessions skips the integration problem entirely. No API keys. No vendor permission. It just needs to see what you see and do what you’d do, except faster and without getting bored after invoice 12.

This is roughly what Dassi does in the side panel. You tell it “pull the line items from this PDF into that form” and it does the typing while you watch. If it misreads a field, you correct it inline. The logins are already present because you’re already logged in. And the data never leaves your browser because the agent is running in your browser. We covered the broader category of these tasks a while back. Data entry is the most boring and most widespread of all of them.

The market that won’t admit it exists

Walk into any mid-sized company and ask how the AR team reconciles statements. The answer involves Excel. The answer always involves Excel. Somebody downloads a CSV from one system, opens it in Excel, opens a second tab to look up vendor IDs, and types the matched ones into a third system.

This is not an edge case. This is most of office work.

A small sample of the soul-crushers

  • Copying LinkedIn profiles into a CRM because Sales Nav’s exporter is gated behind an enterprise plan
  • Transferring expense receipts from Gmail into Expensify
  • Updating product SKUs from a supplier portal into Shopify
  • Moving job applicant data out of a careers inbox and into an ATS
  • Reconciling Stripe payouts against QuickBooks transactions

Every one of these is a browser task. And every one involves two or more tabs the user is already logged into. None of them need a $50k RPA contract. They need something that can look at a screen, read it, and type into another screen.

What I keep thinking about

The salesperson with 90 minutes a day of HubSpot toil makes $140k. Her company is paying a senior IC roughly $50/hour to do work that, if you described it to a 19-year-old in 2005, would have been called “secretarial.” Nothing about the dignity of the task has changed. Only the salary has.

So install Dassi, let it watch you do one of those tasks twice, and have it run the third one while you’re in a meeting. The data entry job will still exist on Indeed next year. But yours does not have to.