A Lead-Qualification Agent Built Without Developers. It Cost Nine Days. Read the Fine Print.
The author built a lead-qualification agent for an enquiry form that scores incoming leads against a rubric in Airtable, drafts replies in her tone, and flags high-value enquiries for a phone call. She emphasizes running the agent for two weeks while logging every mistake, then removing human checkpoints one task at a time. The guide includes free calculators, templates, and courses from a single library.
The principle here is graduated autonomy, the practice of granting an AI agent independence incrementally rather than all at once. The mechanism is trust calibration: you measure failure rates under supervision, identify error patterns, and only relax oversight where performance justifies it. The lesson is that deploying AI agents without a supervised probation period is not efficiency. It is negligence dressed as confidence.
Lilach Bullock, writing at lilachbullock.com, built the agent for her own enquiry form using Airtable as the scoring backend. She reports losing nine days to a single mistake she does not detail but insists most guides skip.
- Create a free Airtable account and build a simple scoring rubric with five criteria, each scored 1 to 5, for evaluating any incoming request or message you receive. You should have a working scoring sheet.
- Open ChatGPT or Claude and paste your rubric along with a sample enquiry, asking the AI to score it and draft a reply in your preferred tone. You should receive a scored response and a draft you can compare to your own judgment.
- For two weeks, run every real enquiry through this manual process and log where the AI scored wrong or drafted something off-key. You should accumulate a record of failures that tells you which tasks the agent can handle alone and which still need you.