GPT Sol & Luna in Make: Building an AI Triage Workflow with Maia and Human-in-the-Loop Approval

In this experimental build, Harrison hands the wheel to AI and sets a timer to see how far Maia can take him. Using just two prompts, one for a Notion database and one for a Make scenario, he builds a complete inbound request triage system. A faster model sorts incoming requests and answers the routine ones, a stronger model steps in only for complex or risky ones, and a human approves everything before it goes out.

🔍 Inside the Build:

* Judgment vs. Precision: Harrison explains his rule of thumb: use AI for judgment-based calls, and use Make’s automation tools for deterministic, surgical tasks like creating an invoice.
* Database First: Learn why most builds should start with the database as context, as Harrison generates an “AI Triage Log” in Notion with properties for category, urgency, risk, handled by, latency, and tokens used.
* Prompting Maia: Watch Maia build the full scenario skeleton from a single prompt: a custom webhook trigger, a classification step, two routers, a response log, a note, and an email.
* Testing Without Postman: See how Maia detects the webhook structure and sends example payloads for you, so you can test straight from the scenario builder.
* Real Requests, Real Tests: Harrison runs six realistic inbound requests (from simple office-hours questions to a client portal error) through the scenario and checks which model handled each one in Notion.
* Real-Time Debugging The first test doesn’t work as intended. Follow Harrison as Maia investigates, iterates, and gets the scenario passing.
* Proving the ROI: Harrison asks Claude to estimate the return on the build, showing how to turn an abstract automation into a clear business case.

Harrison’s mission is to make automation practical and understandable. This video is a great watch for agency owners, operations teams, and no-code builders who want to automate admin-heavy inbound workflows without starting from a blank canvas.

🛠️ Featured Stack:

* Maia (Scenario Building & Testing)
* Make Webhooks (Inbound Requests)
* Notion (AI Triage Log & Approval Queue)
* OpenAI Models (Fast Classification + Stronger Escalation)
* Router & Email Modules (Branching & Human Approval)
* Claude (Build Planning & ROI Estimate) 🔗 Make


https://www.youtube.com/watch?v=FFI51h3qMaE&utm_source=dlvr.it&utm_medium=blogger