CUA vs web scraping — different categories
Classic web scraping sends anonymous HTTP or browser requests to public pages. You only get what a logged-out visitor would see. You pay per request, you can run many at once, and responses come back fast.
A Computer Use Agent works the opposite way: the user grants it permission, and it logs in as that user. You pay per task, you can run only a few at a time (each task needs its own VM — a throwaway virtual machine), and each task is slower (30 seconds to several minutes). The legal picture is clean because the data belongs to the user.
A useful mental model: CUAs are "Plaid for any website" — they bring the open-banking pattern (the user gives permission, then data is pulled out in a structured form) to portals that offer no public API. Think utility bills, bank statements, payroll exports, insurance claims, tax filings, or e-commerce backend orders.
When each one wins
Use a CUA when: the data sits behind a login the user owns; the portal has no API; the job needs MFA, step-up authentication (an extra identity check mid-session), or human-style clicking around; or you need occasional, small-scale retrievals (say 5 documents each for 200 users).
Use traditional scraping when: the data is public (e-commerce listings, search results, social media, news, real estate); you need fast responses (under a second); you need to run many requests in parallel (100+ at once); or cost per request matters (scraping is 10–100× cheaper for the same data when both approaches work).
For 100k items, scraping might cost €20–€100 on Scrappey. Running 100k CUA tasks could cost $5,000–$100,000 depending on the platform. That cost gap is exactly why these are two different tools, not rivals.
The market in May 2026
Anthropic Computer Use — a direct API that drives the real host machine with raw mouse and keyboard actions. Best for building custom agent pipelines. It scores 56% on WebVoyager (a benchmark of real web tasks) because it operates full desktops with all their mess, not stripped-down browser-only VMs.
OpenAI Operator (CUA) — a hosted product with browser control built in; it scores 87% on WebVoyager in controlled environments.
Skyvern — open-source (YC-backed) and driven by a Vision-LLM (a model that reads the screen as an image). It scores 85.8% on WebVoyager and is strong at invoice retrieval, job applications, government forms, and insurance quotes. Available both cloud-hosted and self-hostable.
Browser Use — the leading open-source browser-only agent at 89% WebVoyager, with 78k+ GitHub stars. Plug in any LLM and run it locally or self-hosted. It supports OpenAI, Anthropic, Gemini, and Ollama for local models.
Deck — managed VMs with a credential vault and SOC 2 compliance, positioned as "Plaid for any website" with 100k+ utility provider integrations.