Recoleta
GitHub
Source note

Polsia raised $30M; source map: fake ARR, dead users, god-mode over your company

Authors
not-chatgpt
Source
Hacker News
Published
Collection
Software Intelligence
AI Company BuilderAutonomous AgentsArr Due DiligenceSource Map ExposureHuman In The Loop
Open originalMarkdown source

Summary

The article claims Polsia’s $10M ARR and “fully autonomous” AI company-builder pitch do not match its own public API and shipped source map. It frames the case as technical due diligence on revenue quality, churn, human operations, security, and privacy.

Problem

  • It targets a claimed autonomous AI company-builder that raised $30M while reporting about $10M ARR and 120,000+ companies.
  • The issue matters because investors, users, and customer companies may rely on recurring-revenue, automation, and control claims that the article says are contradicted by public data.
  • It also reports exposed operational telemetry and owner PII for showcased companies, which creates privacy and trust risk.

Approach

  • The authors query Polsia’s public live dashboard API and compare the marketed ARR number with subscription MRR, churn, cost, company counts, and revenue buckets.
  • They reconstruct the public production source map and identify 1,355 source modules, including admin and team-economics UI.
  • They inspect source and API fields for human QA labels, admin users, impersonation, SQL access, credits, and kill or override controls.
  • They probe generated company pages for checkout or payment paths and query public dashboard endpoints for showcased fund companies.
  • They compare their findings with dated third-party snapshots and hands-on user reports.

Results

  • ARR: Polsia’s public dashboard shows a headline $9.70M annualized number, while the article says only about $4.63M is subscription revenue; the rest includes one-off packs ($1.97M), ad-spend pass-through ($1.93M), 1-hour boosts ($0.80M), and user-company payments ($0.36M).
  • Churn: paid churn is reported at about 48% per month, implying about 0.04% of the base survives 12 months; the article estimates durable annual recurring revenue rounds to about $0 under that churn.
  • Unit economics: daily AI cost is listed as $7,344 against about $12,887 per day of subscription run-rate, so compute consumes about 57% of subscription dollars before human ops and infrastructure.
  • Usage: the API reports 118,683 total companies and 7,437 active companies, an active rate of about 6.3%; the article also says all 16 showcased fund companies report $0.00 revenue.
  • Autonomy and control: the public source map contains 1,355 modules, with 464 reconstructable into a running app, and the article identifies human QA labeling, operator logins, admin impersonation, production SQL access, and company halt or override controls.
  • Privacy/security: for 16 fund companies, an unauthenticated endpoint allegedly returned owner name, Twitter handle, balances, 18-agent rosters, per-agent costs, execution timestamps, and a payload field showing public_dashboard_enabled as false.

Link