but.. but they did this in one week

Описание и применение

Использование LLM‑агентов ускорило разработку, но привело к потере владения кодом и навыков отладки. Автор предлагает планировать работу по фазам, вести заметки, читать ошибки самостоятельно и сохранять часть кода под собственным контролем.

Задача

Relying on LLM agents for rapid development leads to loss of developer ownership, debugging skill erosion, and time‑management challenges due to API limits.

Результат

The AI‑driven workflow delivered the product quickly, but developers lost ownership and debugging proficiency, leading to frustration and reliance on the model for fixes.

Рабочий процесс

• 1. Plan work in small, phase‑wise chunks before prompting the agent. 2. Write handover notes after each phase to capture progress and context. 3. Read and understand error traces before asking the agent to fix. 4. Treat debugging as a primary skill, not just code writing. 5. Schedule heavy phases immediately after the agent’s reset time. 6. Keep at least one fully understood part of the codebase (e.g., deployment, DB).

Доказательства

make phase wise plans before you write a single prompt. not "build the app"

write handover notes after every phase

read the error yourself first

debugging is the real skill now