Thoughts on enterprise AI, leadership, and the systems behind better execution.
I write to make my thinking clearer.
The topics usually begin with something I observe in the field: an AI project that looks impressive but cannot survive real operating conditions, a workflow slowed by disconnected systems, a leadership pattern that strengthens or weakens a team, or a widely accepted assumption that deserves to be questioned.
The goal is not to add more commentary to an already noisy market. The goal is to make complex ideas useful.
The whole market has been selling advice as if it were the product. It never was. It was a feature. The product is work completion, and that reframe changes how you buy, measure, and build enterprise AI. The final part of the Expensive Advice series.
A chatbot answers. An execution engine finishes work. The gap between them isn't a better model, it's architecture: shared memory, business context, permissions, and connection to the systems where work lives. Part four of the Expensive Advice series.
In most AI-powered companies, the person is still the runtime, holding the state and connecting the systems by hand. That's the real ceiling on every AI ROI number you've been disappointed by. Part three of the Expensive Advice series.
Every vendor says context improves accuracy. That's the small version of the idea. The bigger one: context exists to eliminate the human orchestration that happens after the answer. Part two of the Expensive Advice series.
The signal that costs you the number is almost always already in your data. It just never shows up anywhere you are looking, in time to matter. Why the systems don't talk, and what actually fixes it.
Every few years someone pitches a giant CRM replacement, and every few years it stalls. The way out isn't a bigger rip-and-replace. It's smaller: move one workflow at a time until the CRM quietly stops being the center of work.
Every vendor shows you how fast their model answers a question. Almost nobody asks the one that determines ROI: who still has to do the work after the answer appears? Part one of the Expensive Advice series.
The AI benchmarks everyone quotes measure single-user tasks in a clean room. Enterprises are the opposite. So we built an open benchmark for that reality, and running the same model, one system hit 94.3% accuracy while the other managed 63.6%.
Most "AI for ITSM" is a suggestion engine bolted onto a 15-year-old ticket architecture. This is what actually moves the needle, why so few tools can do it, and the numbers from the field.
Half of a seller's time goes to process, not selling. The gap between a signed deal and booked revenue is not an effort problem, it is an architecture problem, and it is quietly costing you deals.
A running series on what separates building a team from building an enterprise. Talent gets you started; systems, rhythms, and clarity are what make great work repeatable. New entry each week.