Extended checklist: your first week with an AI agent

Day 1 — pick a pain and talk to the agent

Wrote down 5 problems I understand better than most (work, city, hobby) Picked one: I can explain it to a stranger in a minute Opened an agent (Claude, ChatGPT, Lovable, Base44) and described the outcome, not code: who the user is, what pain, what 'done' looks like Asked the agent: 'what's missing from my description?' — and answered its questions

Days 2–3 — prototype and word-based edits

Asked for the smallest working version: one feature, one screen It works by midnight (doesn't have to be pretty) Gave edits in words: 'this part is awkward, do it like this' — didn't touch the code myself Saved the conversation with the agent — that's my documentation

Days 4–5 — real people

Showed the prototype to three real people from the target audience Wrote down verbatim what they said (not what I wanted to hear) Handed the agent the full list of fixes in one message

Days 6–7 — routine and numbers

Listed my weekly routine and handed the agent at least one task for good Used the calculator to see what an hour of my routine costs Decided what week two is about: improving the product or finding first users

Phrasings that work

'Build the simplest version that solves X for Y' — instead of 'write an app' 'Here's what users said: … What should we fix first?' — agents are good at prioritizing 'Explain what you did as if I'm non-technical' — that's how you learn 'What are the 3 biggest risks in this plan?' — the agent will honestly find weak spots

What not to do

Don't ask for 'perfect' — ask for 'works today' Don't learn to code 'first' — the agent understands plain language Don't build 'for everyone' — one pain, one audience, one result