EUC Forum Winter Meeting 2024 — November 26
Brian gave this keynote at the EUC Forum's Winter Meeting in London — his first UK appearance since before COVID. He set out ten specific reasons why AI agents will enter the enterprise through the same Windows desktops human workers already use, rather than through purpose-built native integrations.
Not all "AI" means the same thing
Brian opened by splitting "AI" into four distinct categories that get conflated constantly: AI research (advancing the underlying science), consumer AI (LinkedIn prompt guides and personal use), operational AI (fraud detection, patient care, case law — AI embedded in a company's core business), and what he calls workplace AI — the AI layer that sits on top of the digital workspace as employees adopt tools like ChatGPT and Copilot on their own. That fourth category, workplace AI, is what the rest of the talk is about.
The building blocks of an AI agent already exist — on the desktop
To act on a worker's behalf, an AI agent needs app integration, application runtimes, a security context and guardrails, a user interface, provisioning, and hardware access — a long list that looks daunting to build from scratch for every application in a company. Brian's point: all of that already exists, already wired together, on the Windows desktop. "The Windows desktop becomes the API for all the AI agents that I might want to use or build in my enterprise."
"I've heard this before" — Windows survives another death sentence
Brian ran through the now-familiar list of technologies that were each supposed to kill Windows (Java, Ajax, Flash, Silverlight, HTML5, SaaS, Chrome OS, mobile apps, the cloud) and noted the difference with AI: earlier waves targeted Windows applications; agentic AI targets the humans using those applications instead. Since the path of least resistance for an AI agent is to use the same tools a human already uses, that difference argues for the desktop mattering more in the AI era, not less.
Ten reasons, the strongest of which:
Minimal disruption. Existing workflows, permissions, and integrations carry over unchanged — the same argument that made Citrix's original 1990s pitch work.
Security and compliance come free. Every control built for untrustworthy humans — DLP, audit logs, session recording, ACLs — already applies to an AI agent using the same desktop, with no new trust model to invent.
Elastic resource management. Unlike a human on an 8-hour shift, an AI agent's VDI session can spin up for 200 seconds at 3am and disappear — the same cloud infrastructure IT already runs, used more efficiently.
The agent doesn't even need to run inside the VM. The pixel stream already flowing to a human user can be cloven off to a screen-archiving tool, a DEX analytics tool, or a workforce-intent model — without resizing or redesigning anything.
Zero new training burden. The way you patch, ring-deploy, audit, and support desktops today is the way you'll support AI-agent desktops tomorrow. Nothing about the operational model changes.
The close: "2025 is the year of VDI"
A running joke Brian has needled about for years — inverted. If AI agents need an always-on, remotely-accessible desktop to operate from, VDI stops being a legacy technology to defend and becomes the actual on-ramp for enterprise AI. His summary of the whole talk: this isn't happening because AI agents are good yet — most are still slow and error-prone — but the direction is clear, arriving in small increments ("drip, drip, drip") rather than one dramatic leap, and the entry point for all of it is the desktop this room has spent 30 years building.
This is a summary. The full talk transcript lives in Brian's second brain: read it on GitHub.