The first AI rollout in most companies looked familiar enough. Buy seats, define acceptable use, run training, create a champion network, ask people to share use cases in a Teams channel, watch that channel briefly look alive and then become one more corporate attic full of good intentions. Fair enough. But the second phase, the “messy middle” of AI adoption, is much stranger. One development team uses Copilot as autocomplete. Another runs agents in tight loops with tests and constant steering. A product owner prototypes real software in five variants. A support team quietly automates the work nobody in the Center of Excellence understood well enough to ask about. Someone figures out how to run something close to a dark factory and proudly tells the customer: I replaced myself.
When the article behind this talk reached the Hacker News front page, the comments were a warning flare for large companies: “I’m not going to selflessly share my productivity gains with the broader company for free” and “developers should treat their personal workflows as trade secrets.” And right in the middle of that hot mess management asks the painful question: what changed because we spent those tokens?
Head of Applied AI at Exxeta.
Robert Glaser is Head of Applied AI at Exxeta. He works at the intersection of AI engineering, business strategy, and go-to-market, helping organizations turn AI ambition into real systems. He previously built and led the Data & AI practice at INNOQ. His work centers on enterprise integration: making modern AI capabilities and agentic systems work with the software, data, and processes companies already have.
