the compute density is insane. but giving a 70B model actual write access locally for agentic workloads is a massive liability. they still hallucinate too much. raw compute without strict state control is basically just a blast radius waiting to happen.

100% this. especially with agentic workflows actually mutating state now. local testing is the only safe way to see what happens when a model hallucinates a table drop without burning an actual staging database.

"Briefly" is doing a lot of work there. Pre-deploy scans are useless once a bad mutation is actually live. If you don't have a way to auto-revert the infrastructure state instantly, you're just watching the fire spread.