Liquidware Chats podcast, hosted by James O’Regan · March 31, 2026 · ~49-minute guest appearance.
Brian’s first appearance on Liquidware Chats. The conversation covers the full arc from VDI Delusion parallels through the cognitive stack to practical advice for enterprises — his most complete public articulation of the cognitive stack framework and the “you can’t automate your way to intelligence” thesis.
The VDI Delusion parallel
James opens by referencing Brian’s 2012 book The VDI Delusion and asks whether AI is the new VDI. Brian draws the parallel: VDI was aspirationally solving the right problems — work from anywhere, stay secure — but delivering the wrong solution. Workers wanted to work their way; the company wanted control. That same tension is replaying with AI. Some companies try to limit workers to what their sanctioned technology can do today, just like VDI vendors did. It doesn’t have to be that way.
Living as a cognitive extension
Brian works remotely, and 90% of his computer interaction is now through AI, used as a cognitive extension. Every meeting gets mined, every idea gets tracked, voice recordings while walking the dog get transcribed and connected to existing threads. He points AI at a folder on his hard drive with everything he’s published, and it mines out core insights, tracks story arcs, rough ideas, and vision concepts. “So now every presentation or every question that comes in, I’m able to sort of call on all of that context.”
Timeline compression
Brian’s seven-stage roadmap article from July 2025 projected cognitive-extension use at “2027 plus,” but he was doing it himself by January 2026. “So six months ago, what I projected would be like two and a half years plus turned into less than six months.” His actual horizon is about two years, but he can’t be sure whether the steps ahead are five years of steps or two months of steps.
The 80/20 framework
Asked where AI fits in EUC specifically: 80% of knowledge work happens inside workers’ heads, invisible to the company, while companies can only see the 20% — emails, documents, meetings. Corporate AI initiatives show disappointing ROI because they only target that visible 20%. Individual workers applying AI to the invisible 80% are seeing transformative results instead.
The cognitive stack
Five layers, like the OSI model: worker at the top, cognitive extension below, then skills, agents, and interfaces and workflows at the bottom. The 80% lives at the top of the stack. The bottom layers — where all of EUC currently operates — are the visible 20%. “You can have the smartest automation layer in the world and it’s not going to sprout up value intelligence out of the top.” You will never automate your way to super intelligence: automations are layer-2 switches. Important, but no amount of layer-2 intelligence creates an application delivery controller.
Applications as compatibility layers
Brian no longer uses Microsoft Word unless interfacing with colleagues who haven’t adopted second brains. He and colleagues share work through GitHub submodules between their second brains. “Teams is an old compatibility layer. Word is an old compatibility layer. To-do lists are an old compatibility layer. Pretty much all of Microsoft Office is an old compatibility layer.”
Security implications
When the previously invisible 80% gets digitized through AI cognitive extensions, the security model changes fundamentally. “We’re talking about tokens instead of packets. We’re talking about MCP data sources. DLP changes completely because you can’t do DLP rules based on regex and scripts when you’re looking at cognitive things.”
Software development as the canary
Brian points to software development as the canary in the coal mine for knowledge work. When Claude 4.5, Gemini 3, and ChatGPT 5.3 shipped in November, the entire coding world changed in two weeks — skeptics to “I don’t write code anymore” in that span. The same thing is coming for all knowledge work, and it will be slow, then fast.
Advice for enterprises
Start by using AI as a cognitive extension yourself: “The first thing you have to do is just start using it and just understand like, whoa, this is really possible today.” Then think through the cognitive stack and understand where value actually lives. The infrastructure foundation still has to be solid — secure endpoints, good performance — but recognize that automating the bottom of the stack is building the foundation, not the endgame. “You’re not automating your way to super intelligence. Like, that’s just not how it works.”
Key formulations
“You will never automate your way to super intelligence.”
“Teams is an old compatibility layer. Word is an old compatibility layer. To-do lists are an old compatibility layer.”
“We’re talking about tokens instead of packets.”
“I’ve always looked at software development as a canary in the coal mine for what’s happening in knowledge work.”
“I don’t think you have two years.”