The pattern that won in medicine
Every successful AI drafting tool for licensed professionals has a visible attestation moment - a step in the workflow where the professional explicitly acknowledges that they have reviewed the AI-generated content and are taking ownership of it.
In medicine, Abridge requires the physician to tap "sign" inside Epic before the AI-drafted clinical note becomes part of the patient record. Ambience Healthcare does the same across hundreds of clinical specialties. Nuance DAX Copilot, backed by Microsoft's acquisition, routes every draft through the physician's sign-off in the EHR.
The pattern is not a legal requirement. It is a design decision that serves three audiences simultaneously: the professional (who needs a clear moment of ownership), the patient or client (who needs to know a human reviewed the work), and the regulator (who needs an auditable trail showing the professional exercised judgment).
What law learned from Avianca
The legal profession's relationship with AI drafting tools shifted permanently in 2023 when Mata v. Avianca produced sanctions against attorneys who filed a ChatGPT-generated brief containing fabricated case citations. Every legal AI tool launched after that case - Harvey, Spellbook, CoCounsel - includes explicit citation verification and a visible "the attorney is filing this" step.
The lesson was not "don't use AI." The lesson was "make the human's sign-off moment unmissable."
What this means for AEC
Sealed engineering work - field reviews under OBC 1.2.2.2, condition assessments, reserve fund studies - carries the same liability structure as a physician's clinical note or an attorney's filed brief. A licensed professional signs, and the signature attaches personal liability.
Yet the engineering profession's tooling has not caught up. Most firms still draft in Word, email the file, and the engineer prints, seals, and scans. There is no workflow-level moment where the tool itself records "this professional reviewed this document and is taking ownership."
Fermito ships an attestation modal - an explicit "Sign and issue this report" step that records the engineer's acknowledgement before the DOCX downloads. The attestation record captures:
- Who signed (the engineer's name, defaulting to the firm profile)
- When they signed (ISO timestamp)
- What they acknowledged (that Fermito drafted the content, the engineer reviewed and owns it)
This is a workflow acknowledgement, not a legal signature - the legal signature is the engineer's ink on the printed DOCX. But the workflow record matters. It means the firm has an auditable trail showing that every report was explicitly reviewed before issuance. When PEO asks how AI-assisted drafting fits into the firm's quality assurance process, the attestation log is the answer.
The moat argument
Defensibility against horizontal AI tools - ChatGPT, Claude, Gemini - is not draft quality. A sharp firm could hand any frontier model five old reports and a firm-voice instruction and get 80% of a draft for $20/month.
The defensibility is the workflow around the signature: attestation, revision lifecycle, regulatory citation tracking, firm template fidelity, and audit trail. A frontier model gives you a draft. Fermito gives you a drafting practice - the workflow infrastructure that a licensed professional needs around the draft to sign it with confidence and defend that signature under scrutiny.
What to look for in any AI drafting tool
If you are evaluating AI drafting tools for your firm, ask this question: does the tool record the moment your engineer takes ownership of the output?
If the answer is "the engineer exports a file and that's it," the tool has not learned the lesson that medicine and law already absorbed. The attestation boundary is not optional. It is the design decision that separates a drafting tool from a liability.