Data and research
Original research across sites: who blocks cloud agents, which bot protection they use, latency and cost.
Benchmark research
Bot protection on 29 websites: vendors named in our notesAcross 29 sites in Wield's registry, notes name Akamai on 6, PerimeterX/HUMAN on 2 and AWS WAF on 1. Cloudflare sits in front of 4 more without challenging.Browser agent cost per task: Claude Sonnet 5.5 on 23 real sitesA Claude Sonnet 5.5 browser agent cost a median of $0.15 per run across 23 real-site tasks ($0.01 to $0.59), versus $0 model cost for a Wield recipe run.How many requests does a task take: Wield recipe vs browser agentA Wield recipe sends 1 to 4 requests per task. The browser agent's per-task median is 204 requests, up to 924 on bahn.de, including page assets and trackers.How websites deliver their data: 34 read recipes by mechanismOf 34 no-login read recipes in Wield's registry, 15 call a JSON API, 9 read server-rendered HTML, 8 read JSON in the page, 1 GraphQL and 1 turbo-stream.Success rate by site: Wield recipes vs a Claude browser agentWield recipes passed 183 of 199 benchmark runs across 23 tasks. A Claude Sonnet 5.5 browser agent passed 63 of 147. Every task is listed, failures included.What share of Wield recipes need a login? 41 recipes on 29 sitesOnly 7 of 41 recipes on 29 sites need a login (17%). The other 34 read public pages. All 7 login recipes are untested, and the benchmark has none.Which sites block cloud browser agents? Benchmark of 23 tasksIn our benchmark a Claude browser agent in a fresh headless Chromium was blocked on 10 of 23 tasks (every run), while Wield ran on the user's own browser.Wield recipe vs browser agent: how much faster, task by taskOn 13 benchmark tasks where the browser agent passed, a Wield recipe took a median 791 ms per run. The agent took a median 70.6 s. Preliminary data.