The developer is an agent harness engineer

On the delivery side, the developer is not primarily a producer, but the steward of production readiness. Her core task is to build and maintain the agent harness. By this we mean everything that surrounds a coding agent and steers and constrains its work: instructions, tools, automated checks, and feedback loops. The harness is the infrastructure in which all three members of the trio can build safely and productively.

The harness includes:

• Guardrails: fitness functions, linter rules, architecture checks, i.e. quality boundaries enforced by machines
• Skills and subagents: specialized skills and agents for recurring tasks, defined and curated by the developer
• CI/CD pipeline: no generated code reaches production without passing a gate
• Observability, deployment, monitoring: technical responsibility remains with her

To do this, the developer must possess the theory of the program, in the sense of Peter Naur’s Programming as Theory Building. Naur argues that a program is more than its code: the knowledge of why it is built the way it is, how its parts fit together and how it can sensibly be changed exists only in the minds of the people who develop it. When agents generate large parts of the code, code comes into being without anyone building the theory behind it. The developer is the person in the trio who still holds this theory and keeps developing it. Only then can she formulate guardrails that enforce the right things, and recognize when generated code deviates from the architecture.

So the developer is responsible not only for the architectural structures, but also for the infrastructure that prevents agentically generated code from undermining them.