ABB AI Hackathon — accelerating the now

Spent a few intense days at the ABB AI Hackathon with the Microsoft team in 2025, working on real industrial AI use cases — not the demo-ware kind, the we’ll-actually-deploy-this kind.

The most useful realization, that I’ve now repeated many times in customer rooms:

The hard part of enterprise AI isn’t the model. It’s the boring layer — auth, retrieval, eval, audit, and the integration into systems that have outlived three CEOs.

Azure brings that boring layer. The work was about plugging the fun layer into it without breaking the warranty.

What we built

(Specifics under NDA, but the pattern was: domain-specific agent + Cosmos DB grounding + Azure AI Search retrieval + a strict eval harness before any rollout.)

What I took away

• Eval harnesses are the moat. The customer with the best evals wins, regardless of model.
• Industrial AI is a systems problem long before it’s a model problem.
• The shortest path from hackathon to production runs through operations, not capability.

→ If you’re building enterprise AI on Azure and want to compare notes, book a chat.