AI gets way smarter when you stop asking it to solve your big problem and instead break that problem into a sequence of tiny, well-defined jobs it can actually nail.
The story
What I just learned about getting wild amounts of value from AI:
AI doesn’t actually want your “big important problem.” It wants a pile of tiny, boring, well-shaped jobs.
When I toss it “Help me build this app,” I get vibes, hand‑waving, and code I’m scared to run.
When I slice that same goal into jobs like:
– “From this description, list constraints + success criteria.”
– “Given those, propose 3 architectures and compare them in a table.”
– “Given chosen design, outline modules, files, and responsibilities.”
– “Now implement just module A, with tests.”
…suddenly the model looks ten times smarter. Because I stopped asking it to be my cofounder and started treating it like a set of specialized tools.
The pattern that clicked:
– Think in jobs, not chats.
– Separate requirements → design → scaffolding → implementation → critique.
– Tag each step as vibe [“explore weird ideas, no code”] or spec [“no creativity, just tighten and formalize”].
– Make every step observable: bullets, tables, test cases, explicit “out of scope.”
– If I don’t know how to break it down, I literally ask:
“Propose a 4–7 step workflow with inputs, outputs, and what ‘good’ looks like.”
The orchestration [my checklist] stays dumb. The individual prompts do the heavy lifting. It feels less like chatting with a magic brain and more like running a clean little pipeline I can reuse, refactor, and upgrade.
That’s the real “AI skill”: not better mega‑prompts, but better decomposition.
Recent Comments