Giving agents the goal, not the prompts

While rebuilding this portfolio, I stopped working in the chat box. I pointed to the portfolios I wanted to measure up to, and went to sleep. Two agents took it from there. One changed the code. The other checked every preview against a list I wrote once. I only looked once the basics passed.

At work that means

• One goal, then a review. People should not have to keep prompting until the answer is right. The agent already has the catalog, the policy, and the person's history.
• Agents check each other. A procurement request goes to the agents that own sourcing, policy, and budget. They come back with one requisition ready to confirm.
• A fixed list catches extra work. Anything I didn't ask for gets flagged instead of shipped. Taste still comes from me: where an animation should focus, or whether a page needs motion at all.
• The person is the final reviewer. They see the best answer the system could find, and they decide. They ask only when something is truly missing.