Resources
Human-in-the-loop AI.
The agent drafts. You decide. That is the product, not a disclaimer at the bottom.
- Fixed scope
- Human approval
- Built around your tools
- Documentation you own
Quick answer
Human-in-the-loop AI means the agent drafts and a named person decides. I do not ship fully autonomous no-review agents. If it can email a customer, write a CRM row, move money, or use your name, a named person approves it. That is the product, not a disclaimer at the bottom.
AI prepares the work; people own the judgment. If I cannot name the reviewer, I will not build it. If the reviewer stops reviewing, I stop the build.
Decision framework
Put a gate on anything that leaves the building, changes a system of record, spends money, or looks like a promise. Let the agent read and draft those steps. Do not let it send or write until a person says yes.
Leave the agent more room on work that is internal, reversible, and already ruled: sorting a pile, pulling a number you already trust, finding a file. Even then, a person should be able to see what happened. Docs you own are part of the loop — the runbook says who reviews what.
This is why the homepage diagram is Inbox → Agent → Human approval → CRM, and why the permissions table is can read, can draft, cannot send, can write after approval. The sample Audit uses the same gates.
Example · EXAMPLE / COMPOSITE
Inbound message. Agent extracts contact and drafts a reply plus a CRM note. Owner approves. Then, and only then, the reply can be sent by that owner and the CRM can be written. Same shape for a dealer group and for a small professional-services shop. No fake client name. No hours-saved figure.
Walk the examples: follow-up, inbox, CRM, reporting, documents, admin prep.
When this applies
It applies to every offer I sell. Fit review, Audit, Personal, Business. The $1,495 personal assistant still waits on you for anything that leaves the machine. The quoted Business build still has approval gates on customer-facing or sensitive steps.
It applies if you are comparing me to a vendor that sells “fully autonomous.” I am the wrong vendor for that ask. I am the right vendor if you want the repetitive work prepared without giving AI control it should not have.
When it does not
If you want no reviewer, do not buy this. If your work is a compliance program I do not hold papers for, I will say I am the wrong vendor — I will not sell HIPAA theater. If the useful job is a one-off judgment, keep it with a person and skip the build. When not to automate is the filter.
Cost and risk
Review gates are why the scope stays one workflow and the prices stay published: Fit review $199, Audit $1,250, Personal $1,495, Business typically $3,750–$7,500. I do not staff an account team. I also do not sell the fantasy that the agent can be trusted with the judgment.
The risk of skipping the loop is a fluent mistake at send-speed. The risk of the loop is a reviewer who rubber-stamps. I design for a one-glance review so the second failure mode is obvious. Custom AI engagements commonly run well beyond this range. Effortless Agents is intentionally scoped around one practical workflow at a time.
FAQ
Does human-in-the-loop mean a meeting every time?
No. It means a named person approves the draft or the write. The review should take a glance, not a standing call.
Can the agent still read my inbox?
If you grant that access, yes. Read and draft are different from send.
Is this only for Business builds?
No. Personal setup uses the same rule. You approve anything that leaves the building.
Next step
Credentials stay least privilege. You grant what the workflow needs. You can revoke it. I do not reuse your workflows as public case studies or publish your data. The diagrams on this site are labeled examples, not a client file.
If you want this rule on a real workflow, tell me what you want handled or start with the Audit. Also see AI workflow automation.