One repeating job. One helper. One owner.

What you leave with

• A live helper on one named job — not a sandbox chat.
• A starter list of facts, each marked proposed, accepted, or ask.
• Written permissions and a runbook.
• One teaching session for the people who will check the work.
• Thirty days of fixes after launch.

How the install runs

1. 1.Name the job. Write the repeating work in one sentence. Name the person who can reject a sloppy draft. If you cannot name the job, do not buy extra AI seats or start a setup.
2. 2.Run a Field Survey. Spend two to three weeks looking at the job, the tools you already pay for, and the facts a helper would need. Helixfield prices this at $9,500. It does not include a live helper.
3. 3.Pick the AI tools that fit. Choose from access, data rules, cost, and your preference — OpenAI, Grok, Claude, Gemini, Cursor, a mix, or what just shipped. How we organize the work today is called Loop & Graph. When a better method ships, we use that.
4. 4.Write down the facts. List the facts, rules, systems, people, and open questions the helper must follow. Mark each one proposed, accepted, or ask. Chat history is not memory.
5. 5.Connect only the tools the job needs. Hook up the software the work already lives in. Write permissions down. A helper that can click every app is a liability.
6. 6.Score it on real work. Test the helper on live jobs, not a scripted demo. Missing facts become questions. Invented numbers fail the test.
7. 7.Teach the owner. The owner learns four simple steps: gather facts, do the job, write down what changed, save it. They must be able to reject a bad fact and refuse an outbound draft.
8. 8.Accept the handoff. A named person accepts the setup before it touches clients. Anything that leaves the company stays a draft until someone says send. Thirty days of fixes follow launch.