MSRLab uses AI agents because they expand the number of ideas, implementations, tests, and critiques one researcher can examine. That is experimental capacity—not delegated conviction.
The research director still chooses the question, judges whether a test answers it, interprets ambiguity, and decides whether evidence is strong enough to affect capital. Agents can challenge those decisions, but they do not silently change risk or deploy live money.
This division matters because automation can produce answers faster than it produces understanding. The lab’s standard is therefore human-governed evidence with an auditable trail.