Citrix AI Think Tank · Paris, France · December 4, 2025 · internal presentation with slides, no video recording.
The canonical stump speech, delivered to Citrix’s internal AI Think Tank before the November 2025 model inflection reshaped it. This version lays out the diffusion gap, how AI actually reaches applications, and the workspace model holding the whole thing together — the direct ancestor of the cognitive-stack talks that replaced it a few months later.
The AI disappointment paradox
Opens with the paradox that’s since become familiar: headline coverage from the New York Times, Fortune, and MIT all pointing to disappointing enterprise AI ROI, set against a workforce that’s already quietly reorganizing itself around AI tools on its own. “AI” means a dozen different things depending on who’s using the word, and most of the disappointment comes from conflating them. The reality on the ground is a shadow strategy, not just shadow IT — workers aren’t only using unsanctioned tools, they’re doing real strategic thinking with them. The talk introduces the seven-stage evolution model at this point, walking from Phase 1 (faster search) into Phase 2 (thinking partner).
How AI actually reaches applications
Modern apps get reached through connectors and APIs. Web apps get reached through browser control — the Operator model. Legacy applications get reached through computer-using agents, with OSWorld benchmark progress charted from 12% up to 72% over a short stretch, alongside Claude’s growing ability to manipulate files directly. The section closes with four stages of what “post-application” actually means in practice: AI doesn’t replace applications, it changes how they get reached.
The workspace model
Phases evolve from AI as helper, to AI as colleague, to AI as agent. A workspace has four components that have to travel together: apps and data, identity, security, and context. Workers are wildly diverse in how they actually work, which is why AI adoption can’t be managed top-down — policy has to flex around real variation rather than assume a single “worker” archetype. The talk models this as a three-way interaction between human, AI, and workspace, with UI flexibility and multiple app-access methods layered on top so the same underlying workspace serves very different working styles.
The transition period, and the historical parallel
A horse-and-buggy image frames the current moment as a genuine transition period, not a finished revolution. The direct parallel is the 1990s promise that the web would kill Windows apps — a promise that never fully delivered, for reasons Citrix built an entire company around solving. The lesson carries forward directly: keep doing what already works, and layer AI on top of it, rather than assuming everything has to be rebuilt from scratch. A platform-architecture slide closes the section, showing where AI sits relative to the existing app and workspace layers.
Alternate Universe IT
A reframe of shadow IT: instead of treating unsanctioned worker tools as a compliance problem to shut down, think of them as “Alternate Universe IT” — a parallel, faster-moving version of IT that’s already answering real questions about what workers need, running just outside the sanctioned perimeter. Referencing Microsoft’s “Vibe Working” framing, the core management-philosophy shift is to manage the environment, not the individual apps inside it — a direct statement of the workspace-as-control-plane principle: let AI live wherever it works best, and govern it at the workspace level instead of trying to control every tool individually. A “Today vs. Tomorrow” slide for Citrix closes out the section.