Architectural

AI in Architecture: Tools for Each Phase of Work, and Where People Still Decide

How architects use AI today: tools for concept, analysis, documentation and drawing review, what they can't do yet, and where a person still makes the call.

Isaac Nemzer
Head of Engineering
A curtain wall sill detail open in Pirros, with one condition flagged in orange for review

Architecture firms use AI in five places today: concept images, design analysis, documentation in Revit, drawing review, and finding the firm's own past work. The tools speed up repetitive work and surface options and issues. They don't carry responsibility: a licensed architect still reviews the work and signs the drawings.

How architects use AI today, phase by phase

AI shows up differently at each stage of a project. The useful question for a firm isn't which tool is newest. It's which phase loses the most hours, and how much a person has to check what comes out. For the software each phase already runs on, see our guide to the best software for architects.

Concept and visualization

Image generators and AI rendering tools turn a sketch, a massing model or a written prompt into images quickly. Designers use them to test moods, materials and directions early, and to show clients more than one option. This is the most visible use of AI in architecture and the least connected to the drawing set. The images don't know the code, the structure, or the budget. Nothing in them carries into the model.

Design analysis

AI-assisted analysis tools generate and compare options for site layouts, massing, daylight and energy use against goals you set. They're useful for testing many options early, before a scheme is fixed. The results are only as good as the inputs and assumptions, so someone who understands the building still has to read them critically.

Documentation in Revit

Production is where the repetitive hours are: sheets, views, tags, notes and parameters, and the late changes that ripple through all of them. This is where an AI agent working inside the model helps most. Mira, the AI agent for Revit, takes a plain-language request and does the repetitive part across a whole view or set.

Type "Center the room tags in this view," and Mira moves each tag, and its room's location point, to the room's true center. Rooms it can't center fairly, such as some L-shaped rooms, are skipped and named. Mira shows its plan before it changes anything, and one undo reverses the whole batch.

Mira centering the room tags on a floor plan: the request, its reasoning, the edits and a report of what it skipped.

Mira's other shipped skills follow the same pattern, each a job you'd otherwise do one element at a time:

  • Renumber the details on a sheet after late changes
  • Bring a view's annotation onto the firm template
  • Find and replace text in every note across the model
  • Set parameters on hundreds of elements from a table
  • Flip grid heads, then repeat the change on the other plan views

Users can also teach Mira their firm's conventions, such as a view-naming standard, as custom skills. For a longer example, see how we used Revit AI on a real home renovation.

Firms that already automate Revit with Dynamo scripts can read our guide to Dynamo for Revit on which jobs suit a script and which suit Mira.

Drawing review

AI review tools check a drawing set for the issues reviewers spend hours hunting: callouts that point to sheets not in the set, keynotes missing from the legend, details that have drifted from the approved version, and code requirements. The Pirros Quality Management System checks the set against the firm's own standards and specifications as well as the code, and every flag names the sheet, the location and the standard behind it. Findings are never applied on their own. A reviewer confirms, dismisses or escalates each one.

Review matters most when the set has to stand on its own. On a design-bid-build project, the contractor isn't chosen until the drawings are out for bid, so they have to answer every question by themselves. "The drawings need to communicate the design intent as clearly as possible, leaving as little room for misinterpretation as we can," says Trent Piazzoni, an Associate and Project Architect at Number TEN Architectural Group, which tested Pirros Quality Management in its beta.

At the detail level, Pirros scores each detail on nine factors, six of them checked by AI vision: crossing leaders, spelling, abbreviation compliance, text overlap, leaders to nowhere and text alignment.

Firm knowledge

Every established firm has solved most of its common conditions before. The answers sit in past projects, named for the sheet they sat on, and the people who know where to look are the senior staff with the least time. AI search changes that. Pirros's detail management and family management systems read every detail and family your firm has drawn: details by their geometry and annotation, families by their parameters. A search for the condition finds it, whatever it was named, with the firm-approved version marked. Pirros in Revit puts the same search in a sidebar, so the right detail goes into the open model instead of being redrawn.

The same search now reaches the AI assistants your team already uses. With the Pirros MCP, part of Pirros Connections, someone can ask Claude or ChatGPT for the firm's details and families and get the drawings back, within the limits your admin sets.

The AIA's AI Task Force provides active guidance around the same opportunity: tools that capture and share best practices from past projects strengthen mentorship and collaboration across a firm.

What AI can't do yet

AI can produce options, drawings and flags quickly. It can't take responsibility for them.

  • Make the design call. It doesn't weigh a client's priorities, a site's context and a budget the way the architect does.
  • Interpret code with liability. A tool can flag a code section. Deciding how it applies to this building, and defending that to an official, is the architect's job.
  • Coordinate the team. Resolving a clash between the structure and the ductwork is a conversation between people, not a search result.
  • Know your firm, unless you give it your firm. A general model knows what's common online, not which parapet detail your office stopped using after it leaked.
  • Check itself. Output can look right and be wrong, so every result needs a person who understands it.

Will AI replace architects?

Absolutely not. As with most major technological advances, AI changes how architects work and makes them more productive. AI is not a substitute for architects, especially since the architect is ultimately responsible for the building's design and safety.

Architects are licensed because their drawings carry legal weight: someone has to stand behind the design, coordinate the disciplines, and answer for public safety. NCARB's Futures Collaborative is direct about this. Architects remain fully accountable for work done with AI, and the architect of record is held to the standard of care whatever a software vendor's disclaimer says. The AIA's guidance draws the same line: AI is good at repetitive work that doesn't call on professional judgment, which frees time for the work that does.

Mira is built on that line. It does production work and shows you all of it, your team reviews it the way they'd review anyone's work, and the licensed professional stamps the drawings.

What does change is how early-career staff learn. Drafting, documentation and coordination have long been how junior architects learn the craft, and that work is changing fastest. A firm that hands production to AI needs another way to pass on what its senior people know, which is why firm knowledge belongs on the list above.

Where it breaks at firm scale

One person trying an AI tool is easy. A firm getting consistent value from AI is harder to achieve:

  • Everyone picks their own tool. Prompts, settings and results live with individuals, so the firm can't repeat what worked or see what didn't.
  • Generic tools don't know your standards. Output comes back in someone else's graphics, naming and assemblies, and someone has to bring it onto the firm's template.
  • Nobody can say what changed. Without a record of what the AI touched and who approved it, QA/QC has nothing to check against.
  • Project data ends up in the wrong places. Client drawings pasted into a consumer chatbot can be stored by the vendor or used to train its models, depending on the tool and its settings. That can conflict with your contracts.
  • Pilots stall. A tool one enthusiast likes rarely becomes practice without an owner, firm guidance and a way to share what works.

How to start at your firm

Start small, keep a person in charge, and let your own results decide what grows.

  1. Pick one hour-heavy task. Late renumbering, annotation cleanup, a QA pass before an issue: something your team does on every project and nobody enjoys.
  2. Test it on real work with a small group. Number TEN does this with every new tool: two or three people try it long enough to judge it, and only then does the firm consider a wider rollout. "We're now at a point where we are seeing applications that could genuinely support our work," says Trent Piazzoni.
  3. Keep a person approving every change. Choose tools that show their plan, wait for confirmation and can be undone.
  4. Write down the firm's rules. Which tools are approved, what project data can go into them, and who reviews AI output before it reaches a set.
  5. Check where the data goes. Use the questions below before any project data goes in.
  6. Measure it. Note how long the task took before and after, so the decision to expand rests on your own numbers.

Questions to ask before adopting an AI tool

Ask these before any AI tool touches a live project.

  • Does it work inside the software we use, or do we export, fix and re-import?
  • Does it show what it will change before it changes it, and can we undo it?
  • Does it use our standards, details and templates, or generic ones?
  • Is there a record of what it did and who approved it?
  • Does the vendor train on our data? Where is the data stored, and what security certification does the vendor hold?
  • Who in the firm owns it: setting it up, writing the guidance and deciding whether it's working?

How Pirros helps: AI that starts from your own work

Pirros is the AI project hub for architects and engineers. It helps your firm find, trust and build on everything it has already solved, and its AI works on your projects, not on what's common online.

  • Find it. Search every detail your firm has drawn by what's in it, across every past project.
  • Trust it. Every result shows whether it's firm-approved or reference only, and the Quality Management System checks the set against your standards and the code before it goes out.
  • Build on it. Mira handles the repetition in Revit; you approve the result. Custom skills teach it your firm's conventions, so it follows your standards rather than generic ones.

Your work stays yours. Pirros is SOC 2 Type II, each firm works in a private workspace, and Pirros never trains on customer models or drawings.

The hours AI takes off production go back to design, to clients and to the craft.

See AI work on your own projects

Book a demo to see Pirros find, check and build on the work your firm has already done. Working in Revit now? Try Mira free.

Talk to a product specialist about Pirros.

Set up a time with our team and see Pirros for yourself.

By submitting, you agree to our Privacy Policy and Terms of Service.

FAQ

Frequently asked questions.

Can't find what you're looking for?
Get in touch.

Yes, several kinds. Image and rendering tools help with concept design, analysis tools compare site and massing options, AI agents such as Mira handle repetitive work inside Revit, and review tools flag issues in drawing sets. Most firms use more than one, each for a different phase.

It depends on the phase you want to speed up. Image and rendering tools suit concept work, analysis tools suit early design, an agent that works inside Revit suits production, and a review tool that checks against your standards suits QA/QC. Start with the phase where your team loses the most hours.

Not construction drawings. General chatbots can explain concepts, draft text and generate images, but they don't produce coordinated, buildable documents tied to a model and your firm's standards. Tools that work inside Revit, like Mira, edit the drawings you already have, with you approving each change.

No. AI is taking on repetitive production and checking work, but architects remain responsible for the design and for drawings that carry legal weight. NCARB and the AIA both stress that architects stay accountable for anything produced with AI.

AI tools can generate layout options and plan concepts from inputs such as a site and a program. Turning one into a buildable, code-compliant, coordinated plan still takes an architect, who remains responsible for the result.

It depends on the tool. Check whether the vendor trains its models on your data, where the data is stored, and what security certification it holds. Pirros is SOC 2 Type II, keeps each firm in a private workspace, and never trains on customer models or drawings.

Explore more