AI and the Architect: Designing Human-AI Workflows That Preserve Authorship

What you’ll be able to do

• Identify the authorship boundary in AI-augmented practice, distinguishing which documentation decisions require human judgment to preserve firm standards and which can be safely delegated to AI execution.
• Apply the judgment-dependent vs. execution-dependent categorization framework, plus the operation-type gradient (reposition / model-new / re-dimension), to 20 documentation tasks, using the same classification logic the firm applied across 100+ AI-run tasks on a 190,000 sf healthcare project.
• Evaluate the standards-governance practices required when AI executes documentation: the review model, JSON-inspectable artifacts, and escalation rules that kept the firm’s conventions intact.
• Assess the professional liability and standards-authorship implications of different AI delegation levels, and apply that assessment to your own firm’s AI adoption decisions.

Session outline

1. Opening · 10 min

   The Authorship Test, From a Real Project

   Kari opens with the 190,000 sf healthcare project and two real edge cases that put the authorship question on the table. First, “Frank,” SWAPP’s AI architect, flagged an ADA clearance violation and repositioned a fixture a few inches to resolve it; that single move, if wrong, triggers $10,000–$30,000 in plumbing rework on a concrete slab. Second, the team pulled one modeling task back from the AI mid-stream (modeling the floor slabs and slab-edge plans), deciding that work belonged to humans. Put side by side, they frame the live question: does the authorship line fall on the task, or on the kind of operation the AI performs? A live ADA check runs on stage against a Revit model, with a pre-recorded backup.

2. Section 1 · 12 min

   What Authorship Means When AI Executes the Drawing

   The two authorships at stake when AI executes documentation: legal/professional authorship (the seal, the liability) and standards authorship (firm conventions, drawing identity). How automation has always pressured the authorship chain, and why AI is qualitatively different: the agent makes judgment calls, not just execution calls. The counter-intuitive twist: trained on the firm’s own “good” projects, the AI surfaced mismatches between overall plans and their enlarged plans that the firm’s drafters had been inconsistent about, enforcing a standard the humans had drifted from. Kari brings a live, unresolved firm debate (should general notes repeat on every sheet, or live on a single notes sheet?) that no AI can settle, because it is a firm-authorship call.

3. Section 2 · 20 min

   The Handoff Framework, Applied to 20 Tasks

   The framework rests on one distinction and adds a second axis. Judgment-dependent decisions, where the right answer depends on design intent, client context, or firm convention, belong to the architect (Life Safety drawings; which details to include on a sheet). Execution-dependent tasks, where the output is fully determined by rules, standards, and model data, are safe AI territory. The operation-type axis: even within execution, is the AI repositioning an existing element, modeling new geometry, or changing an element’s dimensions? The firm’s one mid-stream pull-back was a modeling task. Attendees classify 20 tasks drawn from the 100+ the firm actually ran through the AI, then see how the firm classified the same tasks, and where the room disagreed. The disagreements are the valuable part.

4. Section 3 · 12 min

   Standards Governance: The Operating Playbook

   What the firm actually did to keep authorship intact while running AI on a live healthcare project: the review model, unchanged on purpose — a licensed architect or the BIM administrator reviews and closes everything the AI does, exactly as for a human intern; what changed is that generic “how to draft” redlines dropped sharply. The JSON-inspectable artifact rule: every AI decision produces a human-readable record of what was extracted, what changed, and why. The escalation protocol, in the customer’s words: Frank asks instead of guessing — when it hits a condition outside the firm’s documented standards, it stops and flags rather than inventing. Plus the metrics that mattered: decisions auto-applied vs. flagged vs. escalated, the redline drop, and drift from firm conventions.

5. Closing · 6 min

   The Professional Liability Frame & the Authorship Statement

   What the current professional-liability framework says about automated tools, and where it is silent about AI. The documentation-record question: what you need to keep to demonstrate professional judgment when AI executed the work. What AIA and NCARB have said so far, and what remains open. One exercise: draft a one-paragraph authorship statement for your practice, defining where you stand on the handoff line. Q&A.

Key takeaways

• A two-axis authorship framework for AI-augmented practice (judgment-dependent vs. execution-dependent, plus the operation-type gradient of reposition vs. model-new vs. re-dimension), applied to a 190,000 sf healthcare project and to 20 tasks drawn from 100+ the firm actually ran through the AI.
• The standards-governance playbook the firm used to keep authorship intact: the intern-grade review model AI didn’t change, the sharp drop in “how to draft” redlines, JSON-inspectable artifacts, and an AI that asks instead of guessing.
• A professional-liability orientation for AI delegation decisions: what AIA/NCARB have said so far, what remains an open regulatory question, and what documentation practices protect professional accountability in the meantime.

Speakers