AI-Assisted Engineering AMA #2

A second AI-Assisted Engineering AMA with Aleksey Litvinov, covering local models, multi-agent workflows, brownfield development, security, and agent economics.

The conversation explains why a harness cannot make a weak model strong for free, when multiple agents justify their coordination cost, and how observable behavior, completion criteria, and characterization tests can make a legacy service safer to change.

At the organizational level, the episode connects threat modeling, model and tool gateways, task-level budgets, evaluation-based routing, and the real limit on acceleration: valuable product change rather than token, agent, or pull-request volume.