Computer Use Agents

Work

Invoice processing

The canonical business case. Every organisation has a version of it, and the target system almost never has an API.

What happens

  1. Open the invoice PDF
  2. Read out date, supplier, line items and total
  3. Switch to the finance system and open a new entry
  4. Key each field across
  5. Submit, and record the reference against the source document

One of the tasks Microsoft documents for Copilot Studio, which is a useful signal of what vendors think this is actually for. The work is genuinely mind-numbing, the volume is steady, and the target is very often software with no API and no prospect of one. It is also verifiable after the fact — you can reconcile the entered record against the source document, which most tasks on this list do not allow.

Where it breaks

Invoice layouts vary enormously between suppliers, and a misread decimal is both easy to produce and expensive to find. Reconcile totals automatically rather than trusting the run.

Tools that suit this

Copilot Studio computer use Computer use tool inside Copilot Studio. Automates Windows desktop applications and websites for agents, billed per step. Skyvern Vision-driven browser automation aimed at workflows that break conventional selector-based scripts. Browser Use Python library that hands a browser to any LLM. One of the most widely adopted open-source options.

More work tasks

Work Data entry into a system with no API Where computer use most clearly beats the alternative, because the alternative is a person doing it by hand. Work Extracting data trapped in a legacy application Using an agent as a reluctant API for software that never had one.