Most agent projects fail because they start too high: an assistant that runs sales, answers clients and updates the CRM. Too many things can go wrong at once and nobody can say which one did. A good first agent does exactly one thing, and you can check it in thirty seconds.
What it actually does
You give it a name and a company. It looks at the website and at public news, reads — only reads — your email history with that address, and hands back one page. Not a ten-page report. One page, the kind you can finish in the car.
Name and company
The only thing you type
Website and news
What is public about them
Old emails
Read-only
The one-page brief
- Who the person is
- What changed on their side
- Three opening topics
- Two likely objections
Why it stays read-only
An agent that can send email introduces a different class of risk: a wrong message to a real client cannot be recalled. An agent that only reads is free to be wrong — you correct the page and carry on. Write access is something you grant after watching how it behaves, not before.
Your first agent does not have to be impressive. It has to be checkable.
Terms used here
- Read-only access
- Permission to see data without changing or sending anything. The minimum sensible boundary for a new agent.
- Output
- The concrete thing the agent produces. If you cannot state it in one sentence, the agent is doing too much.
Where to start
- 1Take the next five meetings in your calendar.
- 2Ask for the same page for each one, with the same four sections.
- 3After the fifth, cut the sections you never read and fix the rest as a template.
