Is your team using AI without telling you?
Almost certainly. Jakob Nielsen's UX Roundup (June 29, 2026) reports explosive growth in agentic-AI use among non-developers — people who build and run workflows without writing code — and documents a persistent gap between how designers say they use AI and how they actually use it. If it's happening in design teams, it's happening in your leasing pipeline too.
The practical question for a commercial real estate (CRE) operator isn't whether to allow it. That train left. The question is whether an AI-drafted broker email, rent-roll summary, or letter of intent leaves your firm before a human reads it. Right now, at most firms, nobody can answer that with confidence.
This piece argues the fix is governance, not a policy memo. Stage the action; let a person approve it; keep the record.
What did the research actually find?
Two findings from Nielsen's UX Roundup (June 29, 2026) matter here. First, use of agentic AI — tools that don't just answer but take multi-step actions — is growing sharply among non-developers, the exact profile of a leasing associate or analyst. Second, there's a measurable say-vs-do gap: what people report about their AI use differs from their observed behavior, with actual use running ahead of what they admit.
For an operator, translate that directly:
- Your analyst is likely already drafting tenant follow-ups with a chatbot.
- Your broker may be summarizing a letter of intent — the non-binding term sheet that precedes a lease — with a model you've never approved.
- Your survey asking "does anyone use AI here?" will undercount, because the say-vs-do gap means people underreport.
The lesson isn't "crack down." It's that self-reported policy compliance is not a control. What people tell you understates reality, so a rule you can't observe isn't a rule.
Why is shadow AI a governance problem, not a training problem?
Because the risk isn't that people use AI — it's that AI-drafted work leaves the building unread. A polished email with the wrong RSF (rentable square feet), a summary that misstates the NNN (triple-net) escalation, an LOI figure that's off by a decimal — these ship because generation is fast and review is optional.
Training tells people to "be careful." Governance makes carefulness structural. The difference:
- Training asks a person to remember a rule under deadline pressure.
- Governance puts a checkpoint in the workflow so the action can't complete without approval.
Vantrow's principle is "propose, never commit." The system drafts the broker update, the rent-roll summary, the LOI redline — and stages it. A human reads it, approves or edits, and only then does it send. Every step lands on an audit trail. Nothing autonomous slips out the door.
What should an operator actually do about it?
Assume adoption, then design for it. The say-vs-do gap in Nielsen's June 2026 roundup means banning AI produces two outcomes: people ignore the ban, and you lose all visibility into how they use it. Governed adoption gets you the productivity and the record.
Concrete steps:
- Name the artifacts that must be reviewed. Broker emails, rent-roll updates, LOI language, listing copy, anything with a number or a commitment.
- Route those through a system that stages, not sends. Draft freely; approve deliberately.
- Keep the audit trail. Who drafted, who approved, what changed, when.
- Stop relying on the survey. Measure behavior, not self-report.
The goal isn't to slow your team down. It's to let the desk do the drafting while a human keeps the last word — and to know, after the fact, exactly what went out.
FAQ
(rendered below as schema-ready Q&As)