Articles

Anthropic's computer-use API shipped beta Oct 2024. Production deployments 2025-2026 include Claude Code, Browser-Use derivatives, Airbnb, Amazon, enterprise pilots. Narrow well-scoped tasks succeed; open-ended agent workflows still struggle in production.

WebVoyager top-agent 85%, WebArena 65%, Mind2Web 60-70%. Spread is informative: WebVoyager favors short-task skills, WebArena tests long reasoning, Mind2Web tests cross-site generalization. Triangulating gives the most honest capability read.

Four vendors (Vendr, Tropic, Cledara, Spendflo) sell SaaS pricing intelligence. Vendr/Tropic on customer-panel data; Cledara/Spendflo lean on broader scraping plus customer signals. Market converging — same underlying contracts, same buyer profile, similar recommendations.

Browser-agent benchmarks (OSWorld, WebVoyager, WebArena) show top scores 66-87%. Production agent reliability lands at 45-55%. The 20-point gap is structural — benchmarks miss adversarial conditions, cost constraints, sustained-reliability. Discount benchmark numbers by ~20pts.

HiQ Labs lost at trial 2024. Ninth Circuit's public-data scraping doctrine reshaped by Bright Data v. Meta SJ + district rulings. Public-data scraping protected against CFAA; contract and tort claims remain. Scrapers safer on access, more exposed on downstream use.

Scraping vendor SLAs in 2026 cluster into three bands: self-serve at 99.5% (3.6h/month downtime budget), mid-market at 99.9% (43min/month), enterprise at 99.99% (4.3min/month). Structure mirrors AWS rather than SaaS norms. Enterprise credit-back terms remain customer-by-customer.