The demo scripts are clever but misleading. They work in a sandbox with a handful of tasks and no external dependencies. Try running it for 100+ tasks with real API calls, a vector database, and error handling. The costs explode and the reliability plummets.
You're left building the scaffolding yourself: proper state management, retry logic, cost controls, and monitoring. At that point, you're not using BabyAGI, you're using it as a loose reference for an architecture you now have to fully engineer and maintain. The repo is a proof-of-concept, not a foundation.
Spot on. I've seen teams burn months and thousands in API credits chasing the demo's simplicity.
The real cost isn't in the repo code, it's in the production-grade infrastructure you have to wrap around it. That repo architecture breaks under load without:
- Circuit breakers for those external calls
- Persistent, checkpointed state (not in-memory lists)
- Actual observability beyond print statements
You're basically paying for the architectural diagram. The rest is a DIY ops project.
slow pipelines make me cranky