Survey of 396 Fintechs: 64% of 'AI Agents' Are Just Copilots
LolaCronje · reddit · 2026-08-04
A developer reviewed public AI claims from 396 European fintech companies and found that only 143 (about 36%) showed evidence of an AI system taking real actions in production (e.g., moving money, posting ledger entries, closing cases).
Key Insights:
- Marketing Hype: The other 64% are running Copilots and calling them Agents.
- Failure Modes: A Copilot suggests and a human commits (worst case: wasted time). An Agent commits and a human reviews afterwards (worst case: an irreversible wrong action in production).
- Lack of Incident Tooling: The author notes almost nobody publishes what happens when the second kind gets it wrong. They wonder if the tooling for handling agent errors exists or if everyone is quietly reconciling by hand.
More from AGI Musings
- AI Expert Predicts UHI by 2028-30, Urges UBI Discussion as Unemployment Looms — davidpattersonx · 2026-08-04
- July Recap: AI Gets Cheap as Open-Source Catches Up — Dapper-Tale-4021 · 2026-08-04
- Three Flaws of LLM-Generated Peer Reviews: Endless Nitpicking and Lack of Concrete Judgment — Kwangryeol · 2026-08-04
- FutureSearch Exits Beta: AI Forecasting Now Beats Top Human Predictors — HaydnBelfield · 2026-08-04
- LLMs Winning Fields Medals: Biotech Urged to Test AI's Full Potential — arjunrajlab · 2026-08-04
- Structured Content Attributes Command a Premium for AI Search — yangyi · 2026-08-04