Scope GTM benchmark: only 16.6% of AI agents finish database signup across 18 hosts
.jpg)
1,035 sessions; Prisma Postgres 88.3%, Upstash 63.3%; CAPTCHAs and bot blocks kill most runs. Open leaderboard and transcripts published.
Why it matters
Scope’s first database GTM benchmark asks agents to sign up, provision, write a record and return credentials with no human help across 18 hosted databases and 20 model–tool combos (Claude Code, Codex, Cursor and others). Only 16.6% of 1,035 sessions reached a working result; seven products scored zero. Prisma Postgres led at 88.3%, helped by `npx create-db` paths that skip signup. CAPTCHAs alone stopped 373 sessions and bot blocks another 105. The open leaderboard and session transcripts make agent-readiness a measurable product metric for developer tools. Watch whether vendors strip CAPTCHA/OAuth traps or ship try-before-signup flows.
Key facts
- 1,035 agent sessions across 18 hosted databases
- Overall success 16.6%; 7 products scored zero
- Prisma Postgres 88.3%; Upstash 63.3%; Neon/Nile 43.3%
- CAPTCHAs stopped 373 sessions; bot blocks 105
- Announced on X 2026-11-06T20:22:39Z