AI for small civil and MEP engineering firms: where it fits and who signs

Most small civil and MEP (mechanical, electrical and plumbing) firms are past the question of whether anyone on staff uses AI. Someone does. The useful questions are where it belongs in your work, where it doesn't, and how a licensed engineer stays in charge of what goes out the door. This guide covers all three for firms of roughly 5 to 50 people.

Where engineering firms use AI today

The best recent data comes from a survey of 644 firm leaders by the ACEC (American Council of Engineering Companies) Research Institute and BST Global, run in January 2025 and reported by HPAC in June 2025:

• 85% of leaders see AI as essential.
• 81% use it in marketing and sales, but only 36% in design and project delivery.
• 74% expect to hold staffing flat while producing more.
• 68% think AI could automate up to 29% of tasks.

ACEC's own summary of the research (May 2025) says most firms are still at the pilot stage. It names data quality, cybersecurity and professional liability as barriers, and says respondents see AI as enhancing licensed judgment, not replacing it.

A second study, the PKF O'Connor Davies A&E study (Oct 2025, 2024 data), found that 75% of firms use AI tools internally and 73% have an internal AI committee. That sounds much higher than ACEC's 36%, but the two studies measure different things: PKF counts any internal use, while ACEC splits use by function. PKF does not publish its sample size or firm-size mix, and neither study breaks out small firms.

The pattern is still clear. Proposals and marketing went first. The paperwork around project delivery, which is where your billable hours go, mostly hasn't changed.

What the profession says about responsibility

Every board statement we found, and ASCE's policy, agree on one point: AI is a tool, and the licensed engineer stays accountable. ASCE (American Society of Civil Engineers) Policy Statement 573, adopted July 2024, says it plainly: "AI cannot be held accountable." North Carolina's licensing board went further in July 2025 and told licensees to keep records of what they put into AI tools and how they checked the output.

None of these bodies bans AI drafting. They ask you to stay in responsible charge and to be able to show it. This is a summary, not legal advice. We cover each rule in detail in Can a PE seal work AI helped draft?

Where AI fits in a small firm, safest first

Here are the common candidates, ranked by how cleanly a PE (professional engineer) can review the output:

1. Meeting minutes and design memos. AI turns notes or a recording into a draft, and the engineer corrects it. Low risk and easy to check.
2. Plan-review comment logs and response-letter drafts. AI splits a reviewer's letter into numbered comments, tags each one by sheet and discipline, and drafts a response for the engineer to rewrite or approve. The engineering answer stays with the engineer. See How to track plan-review comments across resubmittals.
3. Proposals, SOQs (statements of qualifications) and resumes. First drafts built from your past projects, with each claim tied to the job it came from.
4. Spec sections edited against your firm's master spec.
5. Submittal review against the spec, producing a compliance table for the engineer to check.
6. QA/QC (quality assurance and quality control) pre-checks on drawings.

Items 1 to 3 are mostly about reorganizing text that an engineer then edits. Items 5 and 6 sit closer to engineering judgment, and that is where software vendors are already competing.

Where it doesn't fit: calculations

In a February 2026 CSE Magazine article, engineers at CDM Smith asked six large language models to estimate heating and cooling loads for a small pump station in Boston. Heating answers ranged from 25 to 150 MBH (thousand BTU per hour). Cooling answers ranged from 12 to 153 MBH. The same authors list meeting minutes, design memos, schedules, budgets and code research as better candidates, and they note that "many engineers have yet to see any significant AI integration."

That spread is the strongest argument we know for keeping chat-style AI out of calculations and on documents, where a reviewer can read every word.

The tools you may already pay for

You don't need a new platform to start. Many firms already have Microsoft 365, a ChatGPT or Claude business plan, or Bluebeam. Set up with your templates, a consistent folder structure and clear rules, those tools can handle items 1 to 3 above. Dedicated platforms go deeper on drawings and submittals, but some are priced for larger teams. We compare the options, with published prices, in Choosing AI tools for a 5-50 person engineering firm.

How to start without making a mess

1. Write a one-page AI use policy. List which tools and plans are approved, what may never be pasted in, who reviews what, and where records are kept.
2. Pick one document type. Choose the one that costs you the most staff time. Measure it first; our comment-letter cost worksheet shows how.
3. Keep a record. For each AI-assisted document, note what went in, which tool was used, where the draft is, who reviewed it, and what they changed. This is close to what North Carolina asks for, and it costs little to do in any state.
4. Train the people who will use it. Keep it short and use your own documents, not generic examples.
5. Check after a month. Did it save time? Did reviewers catch errors? Keep it, fix it or drop it.

What nobody knows yet

There is no published data on how many hours a small firm spends per comment letter or per proposal. We looked and found no one who has measured it for firms your size. That is why step 2 says to measure first. If your own numbers say the time is small, the right move is to do nothing.

---

If you'd like a second set of eyes, start a free conversation. Describe one real comment letter or a recent proposal, and we'll show you where AI would help and where it shouldn't.