5.5 — Building the lead machine (Make + Airtable + ChatGPT) 🖥️

Let’s bring the whole course together into one concrete case, end to end, no code. The goal: turn incoming prospects into qualified opportunities, automatically.

🟢 In plain words — ’‘no-code’’ means without writing a program: you assemble existing tools like Lego bricks. Three bricks are enough — a memory (where to store the info), an orchestration (which triggers the steps), a brain (the AI that decides).

The three building blocks (Level 3):

• 🟦 Airtable = the memory (’‘Leads’’ table: name, company, message, score, status).
• 🟧 Make (or n8n) = the orchestration (the workflow that chains the steps).
• 🟢 ChatGPT (or Claude) = the brain (qualify + write), driven by your Level 2 prompt.

The workflow:

1. Trigger — a prospect fills in a form → a row is created in Airtable.
2. Qualification — ChatGPT reads the message and rates the lead hot / warm / cold (structured prompt, 4 ingredients from Level 2).
3. Memory — the score is written back to Airtable.
4. Action — if hot → personalized email; if warm → follow-up sequence; if cold → archiving.
5. Guardrails (Level 4):
   • filter on the incoming message (against indirect injection: a booby-trapped message slipped into data the agent reads — recall from Level 4) ;
   • HITL before sending emails ;
   • minimal Airtable read/write permissions ;
   • execution history for traces.

🔁 The whole course in one image: concept (L1) → prompt (L2) → agent that acts (L3) → secured agent (L4) → machine that makes money (L5). The same idea, raised a notch at each level.

💰 Why it’s worth money: a salesperson who only handles hot leads (pre-qualified by the agent) converts more in less time. Calculating the gain is the subject of the next demo (5.6) and your workshop (5.10).

⏳ Freshness note: the AI building blocks in Make, n8n and Airtable evolve fast. The assembly — orchestration + memory + brain + guardrails — remains valid.

⚠️ Common mistake — wanting to automate everything at once, without HITL, from day one. An agent that sends emails to real customers without review can ship an awkward or off-topic message at scale. You start with a human who validates before sending, then loosen the reins once trust is established.

❓ Check your understanding — In the lead machine, which brick plays the role of the brain, and which Level of the course tells it how to think?

See the answer

The brain is the AI (ChatGPT or Claude): it reads the message and qualifies the lead. What tells it how to think is the structured prompt from Level 2 (the 4 ingredients: role, context, task, format). Without a good prompt, even a good brain answers poorly.

✅ In short

• The lead machine = 3 bricks: memory (Airtable) + orchestration (Make) + brain (AI).
• The workflow: trigger → qualification → memory → action, with guardrails (L4).
• It’s the whole course brought together: concept → prompt → agent → secured agent → business value.

🔎 Going further

Why separate ’‘orchestration’’ and ’‘brain’’ instead of handing everything to the AI? Because orchestration (Make/n8n) is deterministic and reliable: it always triggers the same steps in the same order, and keeps a trace of each run (useful for Level 4 observability). The AI, on the other hand, is creative but unpredictable: you confine it to the one step where you need judgment (qualify, write). This split — deterministic around an AI core — is a recurring pattern of reliable agentic systems.