Resisting cognitive surrender
Cognitive surrender — giving up your thinking to AI — is a real risk when working with these systems. I’ve faced this challenge as I delegate more of my tasks to AI. The solution: own decision-making even as you delegate production.
AI-drawn blueprints
Greg Petroff and I have worked with two architecture firms over the past year. Making technical drawings always seemed slightly beyond what’s possible for current AI, but that’s changing. Still, I can’t help but feel human-legible drawings are steps in a process that will change further. AI changes firm economics, not just production.
Architecting agentic systems for architects
OK, not just architects: this framework applies for all organizations that deal with the built environment, including builders, managers, etc. Like all fields, AI agents in the AEC space work better when dealing with structured data. This paper reviews the literature and argues for ways to link structured and unstructured data.
Architecting multi-agent systems
Three patterns for designing multi-agent systems. Why does it matter? Because orchestrating agent actions so they don’t trip over each other is a real challenge, especially when scaling into production. The key: start small, grow incrementally, validate outputs along the way.
Box’s Intelligent Content Management
A few weeks ago, I highlighted Stripe’s KAI system for shared AI context. At a recent company event, Box announced upcoming projects aimed at the same challenge: bringing together siloed knowledge safely and at scale. The announcement is short on details, but this is a challenge many organizations must solve sooner or later.
Cognitive surrender is the right name for it. The risk is not that the agent gets the answer wrong, it is that you stop checking because it is usually right. Owning the decision means keeping one step that stays yours no matter how good the draft is. For me that step is the final call on anything that cannot be undone.