You’ve heard the pitch where AI lays out the floor plate and clears code review, then gone back to a week of consultant markups and a fee proposal due Thursday while chasing three timesheets nobody submitted.

This guide shows principals and project leads at 5- to 100-person A&E firms what artificial intelligence (AI) does inside a practice today across design iteration, compliance checking, document review, forecasting, and firm operations, and how to start.

What AI in Engineering Means for an A&E Firm

AI in an A&E firm is software that proposes options, checks work against rules, predicts results, or drafts records, with a licensed person reviewing the output before it goes anywhere. Much of the published work on engineering applications of artificial intelligence covers factories, robotics, and process control. The application of AI in engineering that matters to a design practice sits in five areas your team touches every week:

1. Design iteration and generative options
2. Code and compliance checking
3. Document and submittal review
4. Forecasting and resource planning
5. Firm-operations automation

Most firms start on the office side. The 47th Deltek Clarity A&E Industry Study, which covers 896 firms in 2026, recorded where firms apply AI:

• Administrative tasks such as meeting notes: 59%
• Proposals and marketing: 50%
• Extracting data from documents and drawings: 42%
• Scheduling and cost estimating: 11% each

Meeting notes and proposals carry no seal or life-safety exposure, so a bad output costs your team rework rather than a liability claim. Scheduling and cost estimating trail at 11%, and both depend on clean project data. Your first pilot belongs in the records and documents your office already produces, and design-side uses can follow.

Why AI Matters to A&E Firms Now

AI-adopting A&E firms earned $20,000 more net revenue per employee than non-adopting firms in Monograph’s 2026 benchmark, $210,000 versus $190,000. Monograph’s 2026 benchmark report draws on anonymized platform data from 660 A&E firms rather than surveys.

That gap is an observed difference between two groups of firms. The report describes no controls that would show AI caused it, and it doesn’t define an adoption threshold or isolate any tool or feature. It attributes the difference to more than technology adoption. The AI cohort also captures more of the fees it scopes and keeps more staff time on billable work. The same report calls AI an amplifier for firms whose pricing and utilization are already in order.

Use is rising faster than results. Deltek’s 2026 Clarity study put AI and machine learning use at 70% of firms, up from 53%, while only 38% report measurable positive business impact.

AI Speeds Up Design Iteration and Generative Options

Generative tools pay off in schematic design, when your team wants a dozen defensible options before anyone settles on one. Much of what vendors sell as AI here is optimization or rule-driven solving, and knowing which technique you’re buying tells you how far to trust the output:

• Evolutionary optimization: Generative Design in Revit ranks options against goals you set, using NSGA-II optimization, a multi-objective evolutionary algorithm that runs on Dynamo graphs. It has been available since Revit 2021 to AEC Collection subscribers.
• Site and yield solvers: TestFit generates site and building options, and its CEO told Architect magazine the engine is deterministic.
• Floor-plan generators: Autodesk’s AU 2026 recap still places Forma’s Building Layout Explorer in the research stage, not among capabilities ready for project work. Hypar’s AI documentation says its AI doesn’t lay out or fill a whole floor plan.
• Mechanical, electrical, and plumbing (MEP) routing: Augmenta’s platform, generally available since its June 2026 ACP 2.0 release, routes clash-free electrical raceway. Engineering News-Record reported a school project where estimated manual drafting fell from 400 hours to 80, a vendor-provided figure.
• Structural topology optimization: Skidmore, Owings & Merrill (SOM) used topology-optimization algorithms to shape the cross-braced exoskeleton at 100 Mount Street, cutting steel for lateral loads by 10%. ETH Zurich’s Smart Slab used about 70% less concrete, yet its authors found extra material still cheaper than designing and fabricating the lighter slab.

Plan-level layout generation is still early, and the headline savings figures come from vendors or leave out fabrication costs. Treat the output as options to screen, and find out which technique sits under a product before you trust it.

AI Checks Code Compliance and Design Review

Automated checking handles rules you can express as a property or parameter, and it weakens once a rule depends on spatial reasoning. Solibri runs rulesets, configured to project needs, against building information modeling (BIM) models. buildingSMART’s Information Delivery Specification checks that objects, classifications, materials, and property values arrive as specified.

• Large language model (LLM) code lookup: The International Code Council’s AI Navigator is trained on I-Codes from 2015 to 2024 plus a growing set of state codes. UpCodes says its Copilot links answers to sections adopted in your jurisdiction, a vendor claim.
• Permit pre-screening: Seattle’s CivCheck proof of value caught 87% of the completeness issues city reviewers logged, and reviewers agreed with 92% of its compliance calls (CivCheck evaluation report). Removing whole review cycles, the report notes, would need much broader check coverage.

Spatial rules are where the checks failed. In an ITcon design-review study, LLM-written rules scored 100% on parameter checks and 20–60% on spatial-reasoning rules. Model-property values such as fire ratings and door hardware suit a rule checker. Anything that depends on how spaces relate needs a reviewer.

Submittal and request for information (RFI) review has thinner evidence. We found no independent accuracy audit of a named commercial product, and most published figures are vendor or customer self-reports. Bluebeam warns that a low-quality scan can hide real problems and flag ones that aren’t there for at least one check.

AI Surrogate Models Augment Simulation Like FEA and CFD

A surrogate model learns from a set of solved finite element analysis (FEA) or computational fluid dynamics (CFD) cases and predicts new results in seconds. A physics-based solver computes each case from the governing equations. Four factors set the two apart:

• Speed: Orbital Stack returns AI wind results in minutes, against CFD runs of up to 48 business hours.
• Data needs: SimScale requires at least 20 runs to train a model. Ansys says SimAI typically needs 30 to 100 simulation results.
• Range: SimScale’s guidance warns that a surrogate performs poorly outside its training parameters. Conventional building-energy surrogates stay bound to one site and need retraining for a new location.
• Accuracy claims: A 2024 Nature Machine Intelligence paper found 79% of articles claiming to beat standard numerical methods on fluid problems used weak baselines.

Speedup claims often leave out the cost of generating training data. Autodesk’s Forma tutorial treats rapid wind as directional guidance and reserves detailed analysis for documentation. Use the surrogate to narrow massing options, then run the full solver on the scheme you’ll document.

AI Improves Forecasting and Resource Planning

Monograph’s AI uses historical project data and team availability to recommend fees and staffing, and its forecasts show when that work will land. Pipeline Management multiplies each proposed budget by its win probability and spreads the result across the project’s phase timeline, so a $300,000 clinic proposal at 60% contributes $180,000 of weighted revenue in the months each phase is planned.

Capacity forecasting in Monograph’s Resource Planning weights expected lead hours by win probability and sets them against compensation hours per employee. If your structural group wins two of every five proposals, the forecast shows the month when weighted demand exceeds the group’s compensation hours, so you can hire before the work lands.

The forecasts are only as honest as the win probabilities your partners assign and the phase plans on file. An inflated win probability inflates the weighted revenue and shifts the month you hire.

AI Automates Firm Operations and Time Capture

Monograph, used by 16,000+ architects and engineers across 2,000+ firms, applies AI to the administrative work between a signed contract and a sent invoice. Its six AI capabilities cover setup, pricing, staffing, time, and answers:

• Contract-to-project setup: You upload a contract, and Monograph drafts the project’s budgets and schedule.
• Spreadsheet-to-project import: You import milestone and deliverable sheets as a structured project.
• Proposal budget recommendations: Fee suggestions come from historical project data.
• Staffing recommendations: Suggested assignments reflect team availability across projects.
• Smart Time Tracking: Calendar-based suggestions appear in the web timesheet for you to confirm or adjust, and on the mobile app you describe the work in plain language, and it becomes a structured entry.
• Ask Monograph: You ask plain-language questions about project status, team time, utilization, and profitability, and get answers.

Every Smart Time Tracking suggestion is a candidate entry, and it counts only after the person who did the work confirms or adjusts it. The desktop timer records time only after you start it. Smart Inbox routes work signals into the web timesheet as candidate entries, which also wait for confirmation.

A confirmed Monograph entry lands at the phase level against the phase budget, so your team doesn’t reconcile hours to fees in a spreadsheet.

What AI Cannot Sign Off On in Engineering

AI cannot hold responsible charge or seal a drawing. The National Council of Examiners for Engineering and Surveying (NCEES) Position Statement 6.10 says licensed professionals remain ultimately responsible for their decisions and requires independent checks of AI output before implementation. An April 2026 essay in the NCEES Licensure Exchange, by the head of Mississippi’s licensing board, adds that a professional engineer (PE) is responsible for stamped work whether or not AI was used.

The National Society of Professional Engineers (NSPE) Board of Ethical Review tested this in Case 24-2. An engineer sealed AI-generated design documents after only a cursory review, and the client found misaligned dimensions and missing safety features. The board held that sealing without maintaining responsible charge over the AI output was unethical, and that using AI was not itself unethical. North Carolina’s board goes further and expects licensees to be capable of reproducing output independently of any AI program. Licensing rules keep a named person accountable for every sealed drawing and safety certification. AI assists the engineer, and the engineer still signs.

You still need to manage AI liability. Engineering News-Record reported in October 2026 that no AI exclusions yet appear in design-firm policies, while underwriters are asking about supervision of AI-assisted work. You also need to account for data gaps. Building and sensor records are sparse and noisy. A Pacific Northwest National Laboratory data study found only 9 of about 70 buildings had a year of overlapping meter and controls data, so you need to judge cases absent from the training data.

Common Mistakes Firms Make Adopting AI

Four errors undermine AI adoption:

• Unvetted output: The University of Melbourne and KPMG’s 2025 global trust study found 66% rely on AI output without evaluating its accuracy, across all industries. Case 24-2 shows what that costs when the output carries a seal.
• Poor data: The American Institute of Architects (AIA) AI Firm Toolkit treats data readiness as the precondition for AI readiness. A proposal-budget tool trained on projects coded to the wrong phases will recommend the wrong fees.
• Automated judgment: The same toolkit treats life safety, accessibility, final code interpretation, and stamped deliverables as high risk, where AI stays exploratory and a licensed professional re-verifies the work.
• Tools nobody opens: A 2025 BCG survey of 300 U.S. workers at legal, financial, and tax and accounting firms found 19% had access to specialized AI tools and didn’t use them.

All four mistakes start when you buy a tool before you decide what it should measurably change. Without a measure, nobody checks whether the output is vetted, the data is clean, or the tool gets opened. Set the measure and assign a reviewer first, then choose the tool.

How Your Firm Can Start Using AI

Start with one or two use cases and a measure you can check in a quarter. The American Council of Engineering Companies (ACEC) adoption primer recommends tightly scoped pilots led by a tech-savvy champion with dedicated time and a modest budget. It also recommends reviewing AI already built into your existing tools. Work through it in the order the pilot runs:

1. Pick a low-risk, repeatable task such as project setup, time entry, or meeting notes.
2. Check that the data behind it is clean, with consistent phases, roles, and planned hours across recent projects.
3. Assign a review tier, so a project lead approves medium-risk output and a licensed professional signs anything that touches life safety or code.
4. Set a measure, such as hours to set up a project or days to submit timesheets, and record the baseline.
5. Review results after the pilot and expand only what moved the measure.

The AIA toolkit’s experimentation guidance suggests scheduling three to four hours per person per week of non-billable AI experimentation so the pilot doesn’t compete with deadlines.

Firm operations make a sensible first pilot because spreadsheets and disconnected time trackers hide the relationship between project work and firm finances. Monograph connects pipeline, phase budgets, staffing, timesheets, and billing. Dynamic Engineering, a 10-person structural firm in Florida, reports 25% profit growth after moving from Excel to Monograph. Monograph is sold as a subscription. See current pricing plans. Book a demo to test that connected workflow against your firm’s own project structure.

Frequently Asked Questions

Will AI replace engineers and architects?

No, not in the Bureau of Labor Statistics (BLS) projections. BLS projects 6% growth in its civil engineer outlook and 4% growth in its architect outlook. Its own analysis expects no substantial reductions from generative AI in those engineering occupations. For drafters, BLS projects 1% growth and expects computer-aided design (CAD) and BIM tools to reduce drafting employment.

Can a surrogate model replace FEA or CFD?

Not for anything you submit or stamp. Surrogates screen early options inside their training range. Run the full solver on the scheme you select and on any design outside that range.

What does the Engineering Applications of Artificial Intelligence journal publish?

It publishes peer-reviewed research on practical AI methods across all branches of engineering, per its aims and scope. The Elsevier journal is affiliated with the International Federation of Automatic Control, and its topic list emphasizes control and robotics rather than A&E practice.

Where should a small A&E firm start with AI?

Start with operations work your team repeats every week. Project setup and time entry have clear baselines and low safety risk, as do proposal budgets. You already own the data for these tasks. Design and compliance tools come next, once you have a review process that catches errors before they reach a seal.