"Those who cannot remember the past are condemned to build recursive agent loops with no observability." — probably not George Santayana.
Every decade or so, we rediscover a new flavor of the same old problem. In the '90s it was DLL Hell — version mismatches and shared libraries that made shipping anything feel like defusing a bomb. In the 2010s it was Microservice Hell — over-split systems and a dizzying mesh of internal APIs.
Now we're building Agentic Hell: recursive agent loops with no observability, no provenance, and no way to answer the simplest audit question — what did this system actually do? This piece argues that the fix is the same discipline we eventually applied to every prior "hell": observability, versioning, and verifiable evidence.