AI for architecture is usually discussed in terms of renders and concept images. Some of the more practical opportunities sit in the drawing set. A reflected ceiling plan still shows last month’s layout. A finishes schedule was never updated after a markup came back. Someone spends an hour copying a client’s PDF comments into a register. This post looks at why documentation drifts out of step, what AI agents can reasonably check, and how to judge a tool on traceability, consistency and human review.
Where drawing sets fall out of step
Drawing sets drift out of step through ordinary work.
A new revision is uploaded to the shared folder, and the earlier one stays in someone’s local copy. A consultant marks up page three, and the comment lives in an email thread. The finishes schedule quotes a product from a sheet that has since been superseded. A ceiling height note is hard to read, so the question goes to a services engineer and the answer arrives somewhere else. Information ends up scattered across folders, inboxes and registers.
Catching these problems takes a person comparing documents line by line: which specifications still cite the old revision, which markup requests have a recorded response, whether the room schedule still matches the plan. The work is repetitive, often due at the busiest point of a stage, and harder to complete thoroughly as deadlines tighten. That is where AI agents can help.
What an AI agent does in a drawing review
An AI agent works through a defined task using the information and tools it is given. Say the task is “a new revision has arrived”. The agent opens the drawing, finds the specification entries that cite the previous revision, compares the relevant pages and lists what may need attention.
Traditional automation follows predefined rules and workflows, while a chatbot primarily responds to user requests. An AI agent can choose among permitted actions based on the information it encounters, within its set limits. Our guide to AI agents, chatbots, automation and assistants covers the differences in more detail.
In a drafting workflow, that gives three applications.
Flagging inconsistencies. An agent can flag potential mismatches between drawings, specifications and revisions, such as an entry confirmed against Rev B when Rev C is current. It points to where to look. It cannot guarantee technical accuracy.
Reducing manual work. Transcribing markups, preparing registers and drafting follow-up tasks take time but little design judgement. An agent can prepare a first version of each with the source page attached, so the reviewer corrects a draft instead of typing from scratch.
Managing revisions. By comparing the original, the markup and the revised drawing, an agent can flag requests that may still need attention, such as a client comment with no recorded response.
These checks assist the reviewer. Technical verification and architectural judgement stay with the architect.
What to ask of AI for architecture tools
Put these six questions to any tool you are considering.
- Can you trace each output to its source? A requirement or specification should show the file, revision, page and region it came from, and let you open that exact spot.
- How does it handle a new revision? Items confirmed against the old revision should be flagged for review. They should not carry forward silently, and the earlier evidence should stay on record.
- Who confirms? Suggestions should stay separate from your records until a person accepts them. Finishing a task should not count as approving a specification.
- Does it fit your workflow? Check whether you can start from the files, brief or markup you already have, or whether you must rebuild templates and rename files first.
- What does it check, and what does it leave out? A summary that covers one review should not imply it covers the whole project, particularly near compliance.
- How is project data protected? Ask where files are stored, who can access each project, and whether your drawings or the tool’s outputs are used to train AI models. Drawings and client briefs are confidential, so get the answers in writing.
A drawing review in SpaceMinded
Sixty One Steps built SpaceMinded for designers and architects working alone or in teams on residential, commercial and interior projects. Here is how one review round runs.
A client returns a marked-up PDF asking to change the roof covering. The file goes into the project with the requested change, and it is logged in a change register with its source: file, version, page and region. On the Review screen, the request sits beside the marked-up page, with Original, Markup and Revised views when those drawings have been added. The reviewer can confirm the request, edit it, keep it for review or request clarification, and each response is logged. A linked task such as “confirm roofing specification” is tracked separately.
If the assistant drafts a specification from a drawing region, the draft stays outside the specification register, with its source page and what is unclear, until someone accepts, edits, or dismisses it.
When Rev C arrives, anything confirmed against Rev B is flagged “Review needed”. The earlier evidence and decisions stay on record. The handover summary then lists the included files, recorded decisions and unresolved items, and runs named checks, such as whether every request has a source page. It states that it covers only that review and is not a regulatory, code, or compliance approval.
SpaceMinded is designed to work with the drawings, briefs and markups a firm already uses. Projects can begin with minimal information, with owners, deadlines and room details added as needed. For firms with additional review requirements, Sixty One Steps can assess how to adapt the workflow, which may include custom AI agent development.
Measuring a pilot
Start with one workflow on one live project, such as a client markup round through to the revised drawing. Track four measures:
- Time spent preparing the markup register
- Requests with no recorded response
- Time from receiving markups to completing the review
- How much correction the AI-generated outputs need
Where possible, compare the results with a previous review of similar scope and complexity. Record the number of markup requests alongside the time taken, so differences in workload are visible. The fourth measure shows whether the drafts save time or add checking.
Conclusion
AI for architecture is useful in documentation when outputs can be traced to their sources, changes are flagged for review, and important decisions remain with the architect. If your team handles markups and revisions in PDF and you want to see how SpaceMinded would handle a real drawing review, contact Sixty One Steps with one current project in mind.