The near future of work

DanofficeIT — The Future of Citrix · Copenhagen, Denmark · June 17, 2026

Intimate partner event hosted by DanofficeIT (~25 people). Similar core material to the EUCTech 2026 keynote, delivered in a roundtable format with extended Q&A throughout.

Watch (largely similar version)

A mostly-similar version of this talk was recorded at EUCTech 2026 and published as Citrix AI Hotsheet Episode 2:

The DanofficeIT version includes extended Q&A and some formulations that didn't appear at EUCTech.

Slides

Download slides (PDF)

Key frameworks

The talk

The AI narrative flip and diffusion

The AI narrative has flipped from "AI doesn't work" to "AI costs too much" — and both are diffusion stories, not capability stories. Two clocks run at different speeds: capabilities (still climbing) and diffusion (how fast organizations absorb what AI can already do, hitting a wall). Most "AI ROI" disappointment lives on the diffusion clock.

AI-caused congestion: a worker who produces 10 reports instead of 4 doesn't help if the business can only absorb 4. The bottleneck just moves.

The invisible 80%

Emails, documents, and meeting transcripts are maybe 20% of knowledge work — the visible outputs, the "digital exhaust." The other 80% is invisible: thinking, reasoning, judgment, skill, experience. IT has lived entirely inside the visible 20%. AI changes that. AI can now do the invisible parts, making them digital and therefore manageable. EUC's universe just got much larger.

Why Citrix's 37-year pattern applies

Citrix has always wrapped existing technology to give it modern capabilities without requiring organizations to rewrite everything. In the 1990s: giving Windows apps the reach of web apps. Now: giving AI the same access to applications that human workers already have, without rewriting the applications. AI is going to navigate computers. The question is whether it does so through a governed, policy-compliant Citrix session or through an ad-hoc tool with no audit trail.

How AI enters work: the seven phases

Every worker is somewhere on this path. A key framing: you can only see one step ahead. Someone on Phase 1 can see Phase 2, but Phase 3 looks like a different planet. This explains why AI skeptics exist — they're at the phase where the next one still seems visible and everything beyond is incomprehensible from where they stand.

  1. Faster search — one question, one answer. Most of the world is still here.
  2. Thinking partner — longer conversations, loading documents, real back-and-forth.
  3. Cognitive extension — the second brain. Don't take your documents to the AI; bring your AI to your documents. The AI has access to everything in a vault it reads and writes.
  4. Multi-tool agent — the cognitive extension connects into tools: MCP, browser control, computer-using agents. Not automation — extended reach.
  5. Fleet of AIs — multiple AI systems talking to each other.
  6. The pod — the new unit of work: one worker plus their AI fleet, context vault, and skills. Three worker types emerge: cognitive owners, cognitive operators, and cognitive curators.
  7. The published self (optional fork) — publish your context vault so others can subscribe. Brianmadden.ai/mcp is this.

Session recording and agents

You can do all of this with Citrix today, without new features. Create a second user account (e.g., "Brian Madden Robot") with read-only access. Session recording is on for everything the robot does — 100%, always — because the robot doesn't have privacy rights. Workers' session recording has caused scandals (Microsoft Recall, Facebook). The agent doesn't care. App Protection on. DLP on. All of this is in production on existing Citrix infrastructure today.

Token management

Token consumption scales by phase: roughly 100K/day (faster search) → 1M (thinking partner) → 10M (cognitive extension) → 100M (multi-tool agent) → 1B (fleet) → 10B (always-on pods). I used 291 million tokens my first month of cognitive extension.

The job isn't minimizing tokens — it's maximizing economic value per token. Token routing: the same Excel task can cost 200K tokens (computer-using agent driving Excel), 100K (browser automation), 10K (reading the xlsx directly), 5K (Python script), 2K (reasoning in context), or zero (handing it to a human). This routing is an IT governance layer that didn't exist two years ago.

EUC primitives translated

It's a find-and-replace: users, profiles, apps, policy, sessions → cognitive owners, context, skills, agent policy, agent sessions. VDI stays, used by humans and AI. Image management → skill management. App layering → skill layering. Profile management → context management. Group policy → agent policy. Session recording → cognitive observability.

The consulting model transformation

"The days of a consultant coming in and leaving the PDF after a project — those days are dead." What replaces it: a living context vault developed as part of the engagement, which clients' AIs plug directly into. The consulting product shifts from a deliverable at project close to an ongoing knowledge relationship.

Second brain data integrity

A concrete illustration: my AI built a profile on a colleague based on meeting transcripts. Because I only record disagreements — you don't dictate conversations where everyone agrees — the AI had flagged an adversarial relationship with a colleague I'm 99% aligned with. I caught it only because something seemed off, went directly to the file, and deleted the entry.

The mechanism: selection bias in what gets captured creates systematic distortion in the AI's model of the world. If you only talk to your AI about problems and conflicts, it builds a problem-dominated worldview. The fix: file-based storage you can read, inspect, and edit directly.

Key formulations