Coding is solved. Engineering isn't. Your verification loop is the real bottleneck.

The video cuts through the noise around Boris's claim that 'coding is solved.' The author reframes the debate by distinguishing between coding (the narrow act of generating code) and engineering (the full lifecycle from planning to verification). The core argument is that while AI agents can reliably convert well-specified plans into code, they fail at

catching bugs because our systems don't support automated verification loops. The author points to Anthropic's own Claude Code desktop app as a prime example: text truncation bugs slip through because it's too expensive and complex to spin up graphical environments for testing. This isn't a model capability problem—it's an infrastructure and workflow design problem. If

an agent can't easily spin up and test its own changes, it will ship broken code. The author argues that teams must invest in making their repos 'agent-friendly': simpler local dev setups, better CI/CD tooling, and systems that allow agents to verify outputs autonomously. The tension is that many teams are optimizing for agent speed

(generating more code faster) while ignoring the verification bottleneck that creates downstream bugs. Both Boris and Matt PCO are right in different contexts. Boris focuses on the tactical: agents have surpassed average human coders at writing code. Matt focuses on the strategic: agents lack the foresight to manage codebase health or make architectural tradeoffs. The