Postthe second concept i liked is Tracer Bullets that comes from Progmatic Programmer.

the second concept i liked is Tracer Bullets that comes from Progmatic Programmer. most LLMs tend to code horizontally. they build layer by layer: first models, then services, then UI, and only at the end you see if everything actually works. it can feel productive, but the problem is that feedback comes too late. tracer bullets push you to do the opposite: build a small vertical slice that goes through the whole flow. not the final version. not the perfect architecture. just one thin path that works end to end. for example: one screen, one API call, one database operation, one result back on the screen. what i like about this idea is that planning becomes more grounded. you are not trying to guess everything in advance. you quickly see the real problems: missing requirements, unclear data, wrong assumptions, integration issues. after that, it is much easier to plan the next steps. i think this is especially useful when working with AI coding agents. instead of asking an agent to build a big feature at once, i prefer to start with a tracer bullet first. small working flow → feedback → improve. it keeps the agent on track, and keeps me in control of the direction. here is do-work skill example by Matt: https://github.com/ai-hero-dev/cohort-003-project/blob/live-run-through/.claude/skills/do-work/SKILL.md