AI Workshop
Episode 5 of 10Aired

Prompts that work

Task structure, definitions of done and examples — writing prompts that don't need redoing.

Wednesday 4 November at 19:00 Moscow time· 90 min

Aired

The recording is being prepared — it usually appears within two days. Sign up for the next streams so you don't miss them.

What we'll show

  1. Task structure: goal → context → constraints → definition of done → output format
  2. The difference between «make a form» and a task the agent gets right first time
  3. Examples in a prompt: when one example beats three pages of description
  4. Definitions of done the agent can check itself: commands, tests, screenshots
  5. Before/after on prompts sent in by viewers

What you take away

A task template for agents you can copy into every project.

Task template for an agent

  • Yes: Goal: what should change for the user — one sentence
  • Yes: Context: which files, pages, APIs are involved; a link to a similar place in the code
  • Yes: Constraints: what not to touch, what style, which libraries not to add
  • Yes: Definition of done: the commands and checks that must pass
  • Yes: Example: what a correct result looks like (a code fragment, text, a screenshot)
  • Yes: Output format: a plan before code; at the end, the list of changed files and check output

Harness in this episode

A definition of done in the prompt is the bridge to the harness: what you check by eye today becomes an automated test tomorrow.

The full harness episode

What to have ready

  • Yes: Two or three tasks from your own work that an agent did badly

Official documentation