AI drawing review and the engineer of record: who's responsible, and how to document it

AI tools can now run a first-pass review of an MEP drawing set and flag coordination gaps, code issues and missing details before a human reviewer opens the first sheet. For most engineers, the next question is not whether the tool works. It is what using it means for the seal on the set.

This article is general information, not legal advice. For questions about your own obligations, consult your state licensing board, your attorney, and your professional liability insurer.

Download the one-page decision record template (PDF) A blank record with a row for every finding and a column for each field this guide recommends, ready to keep with an issued set.

If you are still deciding whether to use a tool at all, start with AI drawing review for MEP engineers. This page is about responsibility once you do.

The short answer: AI changes the workflow, not the responsibility

What your seal already means

A seal is a personal statement of responsibility, and nothing about AI changes that. Texas is a clear example. Its rules say the seal is meant to "assure[s] the user of the engineering product that the work has been performed or directly supervised by the professional engineer named," and that "upon sealing, engineers take full professional responsibility for that work" (22 Tex. Admin. Code §137.33).

The trade press puts it the same way. Consulting-Specifying Engineer writes that "the engineer of record is fully responsible for the final design of each project," and that PEs should make sure "all stamped work has been reviewed by the stamping engineer of record, regardless of AI use" (CSE, February 26, 2026).

Seal rules vary by state. Texas is an example here, not a national rule, so check the sealing rules where you are licensed.

What engineering boards and societies say about AI

The bodies that have spoken on AI agree on the core point, even though they word it differently.

Two caveats matter. ASCE writes for civil engineers, though the same principle applies to MEP work. And position statements from NCEES, NSPE and ASCE are guidance, not law. The Idaho board's January 2026 letter to licensees notes that PS 6.10 "has not been adopted into Idaho law or rule." What binds you is your own state's statute and rules, so check your board's current position.

Where AI drawing review fits inside responsible charge

A first-pass reviewer, not a decision-maker

The useful way to think about AI drawing review is as a second set of eyes on work the engineer already directs. NCEES describes the professional's role as "overseeing and validating all critical processes."

NSPE's Board of Ethical Review compares responsible AI use to directing a junior colleague: its recommendations "should not be blindly accepted, they should be considered and challenged and the resulting outputs should be understood" (NSPE BER).

The same case quotes NSPE's definition of responsible charge: "being actively engaged in the engineering process, from conception to completion," and adds that "Reviewing drawings or documents after preparation without involvement in the design and development process does not satisfy the definition of Responsible Charge." The implication for AI review is direct. It can support an engineer who is engaged in the work. It cannot stand in for one.

What "verify" means for an MEP finding

NCEES asks licensees to "critically assess AI outputs, performing independent checks to confirm accuracy and reliability before implementation." The Texas board asks them to review outputs "to ensure they are complete, accurate, and consistent with generally accepted engineering and surveying concepts and principles."

For an MEP finding, that means something concrete:

• Open the referenced sheet and find the flagged location yourself.
• Check the cited code section in the edition your jurisdiction has adopted, not the edition the tool assumes.
• Confirm dimensions and conditions against the current architectural and structural sheets.
• Decide, and write down what you decided.

Dismissing a finding is also an engineering decision, and it deserves a reason as much as accepting one does. A dismissed finding with no note looks, a year later, exactly like a finding nobody read.

There is also direct evidence that AI output needs checking. Consulting-Specifying Engineer asked six large language models for the heating and cooling loads of the same 1,000-square-foot wastewater pump station in Boston and got widely different answers, with cooling estimates ranging from 12 to 24 MBH up to 153 MBH. These were general-purpose chatbots, not drawing review tools, and CSE itself called it "not a rigorous testing process." The lesson still holds: an answer that sounds confident is not the same as an answer that is right.

Competence still applies

AI does not extend what you are qualified to sign. The Texas board says licensees should not use AI "to work on a project in which they do not already have established competence through education or experience," and NCEES says licensees "must not use AI to practice outside of their professional competency."

NSPE's 2024 case, recalling its earlier BER Case 98-3, notes that technology "must not replace or be used as a substitute for engineering judgement." In MEP terms: an AI tool that flags sprinkler spacing does not make a mechanical engineer qualified to sign off on fire protection.

The questions firms are still asking

Some questions do not have settled answers yet. Structured AI's post on AI governance in engineering raises three that come up often:

• If an AI tool catches errors that manual review would have missed, does the firm now have a duty to use AI on all projects?
• If the firm relies on AI and the system misses an error, is the reviewing engineer liable for not performing independent verification?
• Does AI-assisted design review change the firm's standard of care? The post notes that professional indemnity insurers are beginning to ask.

None of the board or society documents reviewed for this article answer the first or third question directly. What they do say, consistently, is that responsibility stays with the licensee and that verification is required.

For the rest, talk to your counsel and your insurer. A June 2026 article by the law firm Tarter Krinsky & Drogin notes that "traditional professional liability policies may not fully address exposure arising from AI use" (JDSupra, June 18, 2026).

Why a decision trail matters

What the guidance says about documentation

• NCEES says AI results "should be verifiable, with clear documentation of methodologies, data sources, and assumptions used in decision-making."
• Tarter Krinsky & Drogin write that "Maintaining records of AI inputs, outputs, and decision-making processes is crucial for traceability, dispute resolution, and regulatory compliance," and recommend identifying the AI platform version in project agreements.
• NSPE Position Statement 03-1774 says "accountability mechanisms must be in place to address unintended consequences or errors arising from AI applications" (NSPE PS 03-1774).
• Structured AI, a vendor in this space, argues that every AI-assisted review should record "what was checked, what was found, what the AI recommended, and what the human reviewer decided."

What to record for each finding

A simple decision-trail example

Illustrative example. Names and notes are fictional.

The value is not the AI's finding. It is the record of the engineer's call and the reason for it, which the next reviewer can follow.

Contracts, disclosure, and client confidentiality

Do you need to tell your client?

NSPE's Board of Ethical Review found no professional or ethical obligation to disclose AI use "unless such disclosure is required under [the engineer's] contract," while adding that "ethical principles favor transparency when AI plays a substantial role." Tarter Krinsky & Drogin recommend that project agreements define permissible AI use, the platform and version, and which deliverables may include AI-generated content, so the question is settled before the work starts.

Client data and confidentiality

The Texas board tells licensees to "be aware of how any AI software utilizes and stores data that is provided to it to ensure that the confidentiality of any client's data is maintained." NSPE's ethics board went further in its case: uploading a client's information into an open-source AI tool "was tantamount to placing the Client's private information in the public domain" without the client's permission. NCEES says that "safeguarding data privacy and security is critical" (NCEES PS 6.10), and the Tarter Krinsky & Drogin article adds encryption and access controls.

Before uploading a client's drawings to any tool, get written answers to three questions. Does it train on your files? Does it share data between firms? Who can access each file? For how to evaluate those answers, see how to evaluate AI QA/QC tools, and for Redline's own answers, what happens to your drawings in AI review.

Insurance

Tarter Krinsky & Drogin note that "traditional professional liability policies may not fully address exposure arising from AI use" (JDSupra, June 18, 2026). Talk to your broker about how your policy treats AI-assisted work before you rely on it.

A practical checklist for adopting AI drawing review

1. Use AI only on work within your competence (Texas PAO 71; NCEES PS 6.10 A.2).
2. Treat every finding as a prompt to check, not a conclusion (NCEES PS 6.10 C.1; NSPE BER).
3. Record a decision and a reason for every finding, including dismissals (NCEES PS 6.10 C.2; Tarter Krinsky & Drogin).
4. Know how the tool stores and uses your data (Texas PAO 71; NSPE BER).
5. Check your contracts for AI-use and disclosure terms (Tarter Krinsky & Drogin; NSPE BER).
6. Talk to your insurer about AI-related coverage (Tarter Krinsky & Drogin).
7. Keep up with your state board's guidance, which is still changing (Idaho board letter; NCEES PS 6.10 E.1).

For the questions to put to a vendor before any of this, see how to evaluate AI QA/QC tools.

What this looks like in a review tool

Redline is built around the principle this guidance describes. It runs the first-pass review of an MEP drawing set uploaded as a PDF and flags coordination gaps, code issues and missing details, sorted by severity, location, and code or sheet reference.

For each finding, the engineer can approve, dispute, or add context. Disputing asks for a reason (not applicable, already addressed via RFI, wrong code reference, false positive, or other), which is kept with the drawing. Every approval, dispute and note keeps its author and timestamp, so the next engineer sees not only what changed, but why. That is the decision trail described above, recorded as the review happens rather than reconstructed afterwards.

The product principle is "A second set of eyes. Never a second stamp." The tool keeps the trail; the engineer keeps the judgment and the seal.

The confidentiality questions above apply to us too. Redline files are never used to train models and never shared between firms, and per-user access controls apply to every file.

The judgment stays with you

AI can take on the repetitive first pass through a drawing set. The judgment and the seal stay with the engineer, and a clear record of each decision, including the findings you set aside and why, is what shows that you stayed in responsible charge.

This article is general information, not legal advice. For questions about your own obligations, consult your state licensing board, your attorney, and your professional liability insurer.