Stop guessing. Define the agent’s job.
Clear scope beats clever demos. A practical way to write agent briefs that lead to shippable behavior.
Most stalled AI work does not fail in the model. It fails in the first week, when nobody agrees what “done” means and the first demo is treated as a strategy.
We write the agent’s job before we write the integration. What is the workflow, in the words of the person who does it today? What decision gets better if this works? What will we refuse to automate in v1?
If we cannot answer those in plain language, we do not start the build. Clever demos without a job description become unpaid second jobs.