AI Stance Is Not Binary: Rethinking Human-Agent Collaboration

Summary

Antonin argues that debates about AI use often present a false choice between delegating all programming to an agent and rejecting AI tools altogether. Using a runner training a robot as an analogy, the essay describes three models. In the first, the human only watches and instructs the machine, gradually losing practical intuition. In the second, the human and machine perform the same work in parallel, creating costly supervision and potential burnout. The third model changes the task into a relay: the engineer handles problem framing, architecture, trade-offs, difficult code, review, and debugging, while the AI handles clearly bounded or repetitive work such as exploring alternatives, preparing tests, searching documentation, or implementing defined components. The author says this approach requires tasks that are meaningful but verifiable, continued human practice, and workflows built around checking results. The essay contrasts this position with both blanket rejection of AI and the view that agents should become the default code-generation tool, arguing that code volume is a poor measure of engineering quality. Its conclusion is that effective AI use depends on learning how to divide work and preserve the quality of the handoff between human and machine.