AI in Architecture: Tools, Workflows, and Design Decisions

AI in architecture can support briefing, research preparation, early option exploration, and review coordination. Choose tools by the task, and keep generated images, preliminary analysis, and verified design information clearly distinguished. Tools like fenbs preserve the approved brief, design decisions, and revision context across connected assistants so a new session starts from the actual state of the work.

For an architect, the useful result is a better-prepared decision or a clearer next step. More options do not help when nobody knows which requirement they answer, what assumptions they use, or whether the client has approved the direction.

Where does AI fit in an architect’s working day?

Start with a task whose inputs and review can be described. An assistant might turn approved meeting notes into a draft brief, prepare questions for a consultant, compare text revisions, or organize an issue list. Those jobs support the design process without pretending that a fluent answer establishes technical correctness.

Keep each output attached to its purpose. A summary for internal discussion is different from text issued to a client or authority. Naming the intended use helps the reviewer decide what evidence and checking the output needs.

This article concerns building design practice; AI system design and architecture decisions covers the separate software meaning. For everyday adoption across a practice, the guide to using AI at work provides a broader starting routine.

What should go into the project brief?

Give the assistant the information that changes the next action: project purpose, current stage, approved requirements, constraints, source references, and unresolved questions. Label each requirement by status. A client request, an option under study, and an approved instruction should not appear in one undifferentiated list.

Suppose a small practice is studying a library renovation. The project brief might say the client approved retaining the main entrance, the team is comparing two reading-room layouts, and accessibility review remains open. The assistant can prepare an option discussion without presenting the layout as settled or inventing an accessibility conclusion.

• Approved direction: what a person chose, who chose it, and the reason.
• Current references: the relevant drawing or model revision and where it lives.
• Constraints: requirements with a source, rather than assumptions repeated as facts.
• Open issues: what needs an answer, who owns it, and the next useful action.
• Use boundaries: what the assistant may prepare and what requires professional review.

Keep project files in the approved document and design systems. An AI brain holds the work context that explains which files matter and what remains to do. In fenbs, that context can stay available when a different connected assistant prepares the next session.

How are visual generation and design analysis different?

A generated image can help discuss atmosphere, material direction, or a possible relationship between spaces. It does not by itself establish dimensions, construction feasibility, accessibility, or compliance. Label it by its purpose and avoid presenting invented details as part of the coordinated design.

Analysis tools address a different question: how a defined proposal performs under stated inputs and assumptions. Autodesk’s Forma environmental analysis guidance↗ (opens in a new tab) describes analyses including sun, daylight, wind, and noise. It distinguishes rapid machine-learning predictions used for early exploration from more detailed analyses for further verification.

Check the specific tool’s documentation for the proposed use. Record the geometry, site assumptions, settings, and limitations used in a comparison. Two attractive diagrams cannot be compared meaningfully if they describe different inputs or methods. A preliminary finding is a reason to investigate, not permission to skip the next required check.

Use text assistance to prepare the review questions: what changed, which criterion improved, what worsened, and what needs a consultant’s attention? Ask the responsible designer to verify the answers against the actual files. Keep any numerical claims tied to their analysis source.

Agree the comparison criteria before producing more options. A design team might consider the approved requirements, circulation, maintainability, and issues identified by the consultants, with the responsible people defining the actual measures. Ask the assistant to describe tradeoffs against those criteria rather than pick the most attractive image. Record uncertainty when an option cannot yet be assessed; a blank answer is more useful than an invented advantage.

What professional checks remain with the architect?

The AIA guidance on responsible AI use↗ (opens in a new tab) treats AI as design support and says architects remain accountable for their work, decisions, and representations. It calls for review and validation by qualified professionals, firm policies, protection of confidential information, and attention to intellectual property.

For US projects, confirm the applicable requirements with current authoritative sources and the professionals responsible for the work. A general assistant’s code summary is a research lead, not a project compliance determination. Check the jurisdiction, adopted edition, amendments, project conditions, and review requirements rather than asking whether a design is simply “code compliant.”

Before uploading drawings, photographs, or client material, confirm the firm’s permission and the tool’s data handling terms. Consider what the material reveals about the client, site, security, and proprietary work. Removing the project name may leave recognizable or sensitive information in the file.

The AIA AI Firm Toolkit↗ (opens in a new tab) offers a structured starting point for readiness, literacy, policy, and implementation. Turn the firm’s chosen procedures into instructions staff can actually follow: allowed tools, allowed information, required review, and the person to ask when the boundary is unclear.

Record those human-approved procedures in fenbs Decisions and rules, where connected assistants read rules first. Recording the procedure helps continuity; it does not certify a tool or transfer professional responsibility.

How should you track revisions across people and assistants?

A useful revision summary identifies the source revision, the verified change, its reason, and the checks still open. Avoid “updated to client feedback” when only some comments were incorporated. The next person needs to know which comments remain unresolved and whether the changed work has been reviewed.

For the hypothetical library project, a summary might say: “Prepared a comparison of the two reading-room layouts using the agreed brief. No drawing changes issued. Project architect must check the circulation assumptions before the client discussion.” That is a useful stopping point because it preserves both the completed preparation and its limits.

When a later decision replaces an earlier one, preserve the reason for the change and make the current direction clear. Otherwise an assistant may combine incompatible instructions from separate stages. Keep references to the current model and drawing set rather than asking a chat summary to serve as the design record.

Use a shared task for each unresolved review issue, with a concrete completion condition. “Review entrance proposal against the identified requirements” is actionable; “fix compliance” hides the question and the evidence needed to answer it.

How to do this with fenbs

1. Create a free account and add one project, such as preparing an approved option-review checklist. One person can start free, with no credit card.
2. Record the project lead’s approved direction and the firm’s AI use boundaries in Decisions and rules. Add a glossary for terms whose meaning needs to stay consistent.
3. Add tasks for open design questions and review work. Reference the authorized drawing and model locations without copying sensitive project files into the work record.
4. Connect a supported assistant through Integrations. Ask it to read the AI context first and save Where we left off with what was prepared, what was checked, and what still needs a person.

fenbs does not generate architectural images, edit building models, perform environmental analysis, or check code compliance. Use specialist design and analysis tools for those jobs. fenbs fits when keeping project requirements, approved decisions, and unfinished review work consistent is the missing part.

What is a sensible first experiment?

Choose an internal task using approved, non-sensitive information: organize a brief, prepare a consultant question list, or draft a revision summary. Compare the result with the source material and have the responsible person identify what was useful, missing, or wrong.

Keep lessons about recurring mistakes and update the procedure before expanding to another task. Judge the complete review effort, not just how quickly a draft appears. The AI by role library offers more workflow ideas; your project’s evidence and the firm’s approved process should govern this one.