Full-stack developer shares workflow for managing AI coding assistants through strict scoping and checks

On November 7, 2026 at 15:00, DEV Community's AI section published a write-up by full-stack developer Amine Affif on how he manages AI coding assistants in his day-to-day work. Affif says he works with a code agent every day and argues that what actually changes the outcome is not the tool itself, but what he does before and after the request.

In his workflow, the front end is deliberately restrictive. He compresses the request into a single sentence, treating that as the threshold for a request being ready: if he cannot state what he wants in one sentence, it is not something to hand off yet, but rather something to discuss with people first. He also sets the scope beforehand, spelling out what the agent may touch and what it must not touch, and he defines the acceptance check up front — for example a test, a page to be opened, or a command that has to pass.

On the back end, he verifies the result, including re-reading the diff rather than taking the generated code at face value. Affif explains the reasoning behind these precautions: if the request is vague, an agent fills the gap with whatever seems plausible, and plausible does not necessarily mean correct. The upfront framing and the post-hoc checks are his way of countering that uncertainty. The supplied excerpt ends mid-sentence at the point where he describes re-reading the diff, so the later part of his write-up is not available in the material provided.