A system that combines different kinds of parts costs less, runs faster, and often gets a better answer than a bigger model alone.
Ranjit Kohli, YYZdata Inc. October 22, 2026. Paper 4 of 5.
One of the most useful lessons of my career had very little to do with technology. In the late 1990s I took over product development at Global One, the international carrier owned by France Telecom, Deutsche Telekom and Sprint, and our products were not getting out the door. Marketing would announce something, sales would start selling it, product management was all in, and the product wasn't ready. Customers were frustrated with the delays. Every team was working hard, on everything at once. More effort wasn't going to fix that. Deciding what mattered did.
With management's backing, we set up a new process. At the center of it was a priority matrix, and it was completely transparent. Everyone could see which projects we were working on and which were lower priority, to be touched only with spare cycles and never at the expense of what was needed. We brought billing and provisioning together and mapped how an order actually flowed from one to the other, charted everything, and sent out updates constantly, so both groups understood the whole flow. That was a breakthrough. Then we sequenced the announcement, the orders and the build, so that by the time a customer needed the product, it was ready. Time to market came down a long way.
I think about that job a lot now, because I see the same pattern with AI. The problem at Global One was never effort. It was deciding what deserved the effort, and how the pieces fit together. With AI, the default has become to send every task to the biggest model available, the way every team at Global One worked on every project. Just ask the AI. The fix is the one we used then: decide what really needs the expensive resource, and map how the pieces fit together before throwing more at it.
PDF, 7 pages.
Ranjit Kohli is founder and CEO of YYZdata Inc., whose lead product is Signals, an operational intelligence platform.
Paper 1 of 5: The model is the pipe
Paper 2 of 5: The whole model, every time
Paper 3 of 5: Read first, guess last
Paper 4 of 5: Who answers the question
Paper 5 of 5: I watched the last pipe get overbuilt
© 2026 Ranjit Kohli. Published by YYZdata Inc. This paper may be shared and quoted with attribution.