Practice standard 8 min read

Where AI should—and should not—sit in architecture work

A practical boundary for choosing appropriate architecture AI tasks, protecting information, retaining human responsibility, and documenting review.

Author
Stefan Miller · Founder + CEO
Published and updated
Published August 25, 2026 · Updated August 25, 2026
Share:

Direct answer

Answer

Use AI for bounded research, drafting, organization, comparison, visualization, and automation support when sources and outputs can be reviewed; do not delegate professional judgment, approval, confidentiality decisions, or responsibility for public-facing work to the model.

01

Use AI where review is possible

Choose tasks where the source material can be identified, the output can be inspected, and errors can be corrected before reliance. Good candidates include first drafts, structured comparisons, issue extraction, research organization, controlled visualization, and repetitive data operations.

Define the person responsible for review before the run starts.

02

Protect project information

Do not upload confidential, personal, proprietary, or client information until the firm has confirmed the provider, account, retention, training, location, and access controls. Remove unnecessary identifiers and use approved environments.

Record which model and version contributed when the use is material to the work.

03

Keep professional responsibility human

AI output should not be represented as code compliance, professional advice, final coordination, client approval, contract interpretation, or authorization to build. The RAIC states that AI may assist practice but does not replace an architect's duty of care, judgment, ethical responsibility, or accountability.

Make uncertainty visible and route high-risk decisions to the responsible professional.

04

Build a reviewable practice habit

Maintain a short use record: task, source set, tool, model, prompt version, output, reviewer, corrections, and final disposition. Review recurring failures and update the Prompt or stop using it for that task.

The goal is not to automate everything. It is to create a controlled process that makes useful work faster without hiding how the result was produced.

REVIEW BEFORE USE

Known limitations

  • Professional duties and permitted services vary by jurisdiction, role, contract, and licensure status; this guide is not legal or professional advice.
  • Model behavior, retention, training policies, and security controls vary by provider and can change.

PRIMARY SOURCES

References used for this guide

These guides support Prompt evaluation and do not replace project-specific advice, code interpretation, consultant coordination, or qualified professional review.

Reviewed by archiPrompt Studio · Editorial + product review

Privacy choices

Choose what archiPrompt measures.

Necessary

Authentication, checkout, security, and remembering this choice.

Always on

Marketing tracking and session replay are not in use. You can change this choice at any time from the site footer.