DAIR.AI's Elvis Saravia and creator Alex Volkov debate whether custom verifiers in LLM agent loops can replace manual human code review

Strongly agree with all of the above. We are entering the next era of code, where the model is able to generate correct code for an increasingly large percent of tasks.

Our job is to make sure the model and our systems have the right guardrails, then to run Claude Code + an advanced model + a verifier in a loop and feed it tasks (or, give the model the data it needs to generate tasks), finding and getting rid of bottlenecks along the way.