Skip to content
Dobi

Your first task

Pick something you ask your agent to do more than once: a weekly summary, a code review checklist, a data clean-up.

  1. Do it once with your agent, as you normally would.

  2. Then ask:

    Use Dobi to turn what I just did into a task, and record this run.

    The agent proposes a task: what it’s for, a few example inputs, and how to tell a good result from a bad one. Check them. It’s your definition, not the agent’s.

  3. Next time, ask the agent to use Dobi for that task by name. It runs it, records the output, judges it against your checks, and counts the tokens.

  4. To improve it, ask the agent to try a better version: a different prompt, model, tool or workflow. Dobi compares the versions on the same inputs, so you can see which does the job, and for how many tokens.

The app shows each task’s runs, their results and their cost, once you’re connected.

Token counts come from what your host records. Dobi marks anything it couldn’t see as unknown, never as zero, so a comparison says when it’s incomplete.