EUCtech Denmark 2025 — May 22
Brian gave the closing keynote at EUCtech Denmark 2025 in Billund, to an audience of EUC practitioners and IT professionals. His thesis: AI agents are already inside the building, they're entering the workplace through the same workspace as control plane humans already use, and that pattern is going to last far longer than most people assume.
AI is arriving in drips, not a wave
Brian opened by reframing the "when will AI hit the workplace" question — not a future tsunami, but rising water you don't notice until your socks are already wet. Most people's jobs are made up of dozens of individual tasks, and AI today handles maybe 5-6% of any one person's — not dramatic on its own, but multiplied across an entire company, that small percentage is already showing up as layoffs blamed on AI, well before AI is doing anything close to a majority of anyone's job.
"I am a super-worker"
The personal version of the argument: Brian estimates the volume and speed of what he now produces in his own role — using ChatGPT, the Microsoft stack, and the Google stack side by side — would have taken a team of three people four years ago. Not because his job changed, but because his AI-augmented output did. "I am a super-worker... What am I doing next week, next month, next year? Four people, five people, six people — very, very, very small little steps."
From chatbot to agent, one small trust step at a time
The trajectory he traces: AI starts as a separate app you copy-paste into, then it can watch your screen, then make suggestions, then take small permitted actions while you supervise, then run unattended while you get coffee. Each step is small; the cumulative direction is toward what the industry now calls agentic AI — autonomous action, not just answers. He cites Sam Altman's framing directly: "AI models will soon serve as autonomous personal assistants who carry out specific tasks on our behalf."
Why build a custom agent when the AI can just use your computer?
Brian's own experience trying to build an agent — hunting down APIs, tokens, and cloud plumbing just to automate an expense report — is the setup for his real argument: computer-using agents (Claude Computer Use, OpenAI Operator, Microsoft's equivalents) skip all of that by literally operating your existing computer — moving the mouse, clicking, typing — the same way a human would. On the OSWorld benchmark, human performance sits around 72%; the leading agent scored 1% in March 2024 and 42% by this talk in May 2025. Microsoft's parallel Windows Agent Arena benchmark shows the same trajectory (19% to 52%).
Why micro-apps failed, and why computer-using agents won't repeat the mistake
Micro-apps — narrowly-scoped, custom-built workflow apps for AI agents — never scaled, Brian argued, because every one required bespoke integration work, someone to decide which of an organization's thousands of workflows deserved one, and ongoing maintenance, while shadow use of the real applications continued anyway. The humanoid-robot analogy carries the point: robots are shaped like humans not out of sentiment, but because the world — doors, tools, stairs — is already built for that form factor. Computer-using agents work through the world businesses already have, rather than requiring a bespoke integration for every task.
The workspace, not the app, is what survives
Brian's core structural claim: a workspace is apps, data, identity, security, and context, and for 30-plus years the worker connecting to it has been human. He argues there's no meaningful difference, from an IT administrator's seat, between a human worker and an AI worker connecting to that same workspace — same guardrails, same session recording, same audit trail. When a human closes their laptop lid, an AI worker that needs to keep running just moves to a cloud-hosted version of the same workspace — which VDI already provides, and which a browser-based BYO setup can approximate with a small secured container.
He closes the "this is temporary" objection with a long, deliberately funny list of technologies that were each supposed to kill Windows and Citrix-style remote computing — Java, Flash, Ajax, Silverlight, HTML5, SaaS, mobile apps, the cloud — none of which did. His bet is that AI agents entering through the existing workspace model isn't a stopgap either; it's the durable pattern, because it requires no rebuild and inherits every governance tool IT already has.
His close: "This is happening whether you like it or not... The good news: you're not falling behind. You are preparing to lead this next revolution."
This is a summary. The full talk transcript lives in Brian's second brain: read it on GitHub.