Agentic development uses AI agents to accelerate programming, while keeping human judgment responsible for quality, design, and long-term durability. section 1.0 section 1.3
Use agents as capable assistants, not as autonomous custodians of system design:
- Inspect the surrounding code first; agents often mirror whether it is clean and expressive or tangled and unclear. section 1.2.2
- Treat an incorrect first draft as a design problem, not an invitation to accumulate successive patches. section 1.2.2
- Supply system context, business constraints, and long-term consequences that the agent cannot reliably infer on its own. section 1.2.2
- Review the complete solution for comprehension debt: working code that no one can fully explain, trust, or safely evolve. section 1.3
When a quick fix becomes comprehension debt
An agent generates code inside a tangled codebase. Repeated requests to correct an incomplete result add patches, and the final implementation compiles and passes tests without becoming understandable. Reframing the design around clearer intent—and accepting responsibility for the final structure—avoids mistaking immediate success for durable quality. section 1.2.2 section 1.3
