Adoption is the hard part: 8.5 months of MCP in production at an HVAC company
8.5 months of MCP in production at a 350-person Dutch HVAC company: two adoptions, 18 servers, and what happens once colleagues outside IT start building.
- Level
- Intermediate
- Language
- English
- AI
- MCP
- Agentic AI
- Enterprise AI
- Real-world adoption
Where I'm giving this talk
Abstract
Numbers in the accepted abstract reflect the platform at submission time. Current figures: 18 servers, 114 tools, 128 active users.
Most "AI in the enterprise" stories come from tech giants. This one comes from a 350-person Dutch HVAC company with a small IT team, where AI was brand new.
There were two adoptions. Inside IT, better coding models were never enough: output only really took off once the team moved to a fully AI-native workflow. Outside IT, it clicked the moment project managers and engineers could ask their own data questions in plain language. Behind that: a production MCP platform of 18 servers, built in eight and a half months, across the ERP, BIM models, energy data and more, with the meaning of the data built into the tools.
I will show the real numbers and the shape behind them: what pulled people in, and what quietly stalled. And the strategy that let a small IT team add value fast: scale capabilities, not agents.
Then came the unexpected part: once it worked, colleagues outside IT started building their own AI workflows, and even software.
This is a field report, not a roadmap. You will see what AI adoption actually looks like in an ordinary company, what makes it work, and what to prepare for when it succeeds.