AI for MEP Engineers

Chat: AI as a general assistant

Chat AI is the tool most engineers have already tried — ChatGPT, Copilot, Gemini, Claude. It works as a general-purpose assistant sitting alongside your engineering software. Ask it to draft a specification, summarize a code section, walk you through a research question, translate a client email, build an Excel formula, or explain an unfamiliar term. It is helpful, sometimes remarkable, but it operates on words and images. It does not open your Revit model, does not read your panel schedule, and does not do engineering calculations.

Chat is a productivity tool. It shortens the time it takes to write, research, and organize. It does not do engineering work.

Agent: AI that understands, operates, and generates

Agent AI is the newer, more consequential category. Instead of talking about your project, it interacts with your project. Three capabilities matter, and the line between them is where most of today’s development is happening.

AI that understands the project

An agent can interrogate drawings, models, and project data and answer questions about this building. Not “what is a typical panelboard voltage” but “which panels on level 3 are more than 80% loaded” or “list every fire alarm device without a room number.” It reads the model the way a knowledgeable engineer would — but faster and without missing rows.

AI that operates engineering software

The next step is action. An agent can create circuits, add loads, edit equipment, place families, adjust parameters, generate schedules, and make targeted changes in the live model. The engineer describes the intent in plain language; the agent uses the same released tools the engineer would use manually, and the engineer reviews the result. Autodesk has said explicitly that Revit is moving in this direction, including MCP-powered model queries and element operations. Products like ElectroBIM and FireBIM already ship this today.

AI that generates engineering designs

The most ambitious version is generation: given a set of requirements and constraints, the agent produces some portion of an HVAC, electrical, or plumbing system for the engineer to review, revise, and approve. Several companies are pursuing versions of this — from AI-native MEP platforms to targeted routing and layout tools. Generation shifts the engineer’s role from constructing every element to reviewing, adjusting, and approving proposals.

These three capabilities usually appear together in an agent product, but the balance varies. Understanding where a product sits on this spectrum matters more than the marketing label.

AI for Electrical Engineers

Electrical engineering is where AI-assisted engineering software is most mature today, in large part because the discipline is heavily rules-driven and heavily documented. Panel schedules, circuit loads, feeder sizing, voltage-drop and fault calculations, one-line diagrams, coordination studies, and equipment specifications all live inside strict codes and repeatable formulas. That structure is a good fit for agent AI: the engineer states the intent, the software runs the deterministic calculation, and the AI reviews the result against the code, the project standard, and the rest of the model.

Today an AI assistant working inside Revit can create and balance panels, size feeders and branch circuits, trace the electrical distribution system upstream or downstream, adjust equipment parameters in place, and produce a design-review report on the current model. The engineer sees every change in the live drawings, retains control over decisions, and undoes anything that is not right. This is not AI replacing the engineer — it is AI operating the tools the engineer already uses so the engineer can spend more time on judgment calls.

See the AI electrical design landscape →

AI for Mechanical / HVAC Engineers

Mechanical and HVAC engineering has a wider spread of work: block and detailed load calculations, equipment selection, system layout, duct routing and sizing, hydronic piping, controls, and energy analysis. Some of this work is highly deterministic — a load calculation follows well-defined thermodynamic rules — but other parts, like laying out a system across a floor plate or coordinating with other trades, involve more judgment and constraint-negotiation. Both parts are targets for AI.

Agent AI in HVAC looks like this today: reading the architectural and structural model to establish the envelope, running or checking a load calculation, selecting equipment against a schedule of available families, proposing a first-pass duct or piping layout for the engineer to refine, and reviewing the finished system for coordination clashes and code compliance. Because HVAC design touches so many parts of the building, the AI’s ability to read the whole project — not just the mechanical drawings — is often what makes it useful.

AI for Plumbing Engineers

Plumbing design shares many of the same shapes as HVAC — routing pipe from source to fixture, sizing to code, coordinating with other systems — but with its own fixture information, water and drainage calculations, and sanitary/storm requirements. The routing side benefits directly from spatial AI that can find clash-free paths through congested ceiling spaces; the sizing and calculation side benefits from agent AI that can walk from fixture units through pipe sizes, pressure drops, and vent requirements.

Because plumbing tends to be tightly coordinated with mechanical and structural work, an agent that can query the full model — fixture counts, distances, elevations, adjacent equipment — is often the difference between a tool that shortens design time and one that shortens rework time.

The most important thing to understand about agent AI in engineering is what it does not do. It does not perform the engineering calculation.

The calculation — sizing a feeder, running a load, checking voltage drop, computing pressure loss — happens in deterministic engineering software. Those calculations must be repeatable, defensible, and traceable to the code and standards they came from. AI is not a substitute for that; it is a partner around it. The engineering software does the math; the AI helps operate the software.

In practice this looks like a loop: