CPC AI's Yoon: 3–5 day MVPs come from cutting decisions, not faster code

Yoon's post is unusual in that it spends more space on what he refuses than on what he ships. The 3–5 day figure only holds, he writes, when three things are settled before any estimate: which features are dropped, which parts are bought as managed services, and which files the AI is not allowed to touch. If those are not agreed, he says the days are unpredictable and he will not take the job on that timeline.

The pipeline itself is built around a working production skeleton on day 1 — login, one record created, one list view live at a real URL — after which later screens repeat the same pattern. The failure modes he lists are specific: dependencies added just to make things run, tests that pass but do not guard the main flow, and type holes papered over with 'any'. His fix is to put strict type settings, lint, one E2E test, a preview environment and an explicit no-AI file list into the repository from the start.

On price, Yoon says the lower quotes come from trimming process and rework upfront rather than from cheap labor, and that he prices by task instead of by person-month. The piece reads as a working note from a one-person consultancy rather than a product announcement, and its claims are his own account of his own practice.