AgentBench — Independently verify AI startups' claims

Help buyers validate AI companies' technical claims and customer numbers from public data

Problem

Someone independently checked claims from five AI businesses using only public code, data and official documents; technical claims like a parametric 3D cone being rebuilt rather than stretched were easy to verify, but customer numbers were nearly impossible to confirm. Buyers end up making sourcing decisions from marketing pages alone.

Solution

A lightweight verification service: enter an AI company domain, automatically pull its public product pages, GitHub, docs and APIs, cross-check quantifiable technical claims, and output a verified / unverifiable / questionable list; for customer numbers, aggregate only publicly available signals such as shared customer logos and case-study links without promising hidden data.

Users

Enterprise AI procurement leads, investment due-diligence analysts, AI product managers and cautious developers.

MVP

Auto-scrape public technical claims from a domain; provide reproducible test templates for quantifiable claims; clearly separate verifiable technical claims from unverifiable customer claims; export a PDF brief; no private-data scraping beyond public sources.

Revenue

Per-report pricing; team subscription; API billing for due-diligence platforms.

Why now

The number of AI startups is exploding, making the gap between claims and reality a real procurement risk; an indie developer can focus purely on structured public-data verification without needing massive datasets or licenses.