A New Direction: AI Operating Systems

This Substack started as a place for my thoughts on strategic finance, FP&A, and organizational decision-making.

My work has moved from finance into AI product development and AI scaling and strategy, see my LinkedIn. The questions top of mind have moved too.

Most of my focus is now on how enterprises turn AI capabilities into valuable, dependable work. The model is mostly not the main problem. The real work is around it: operating models, product judgment, evals, engineering practices, governance, change management, cost discipline, and evidence that the solution actually improves something.

That is the new focus of AI Operating Systems.

I will publish posts about AI product management, engineering, enterprise adoption, governance, economics, and the mechanics that connect them. The perspective will stay practical and opinionated, grounded in hands-on experience, research (lots of newsletters and podcasts mostly), and more than 20 years inside banking.

I’m relying heavily on Codex (currently) to create the final posts. As we’re deep into building internal AI solutions at daily work and I believe more in systems than creating outputs, the outcomes are as good as I’ve been able to guide my research and writing system. Each post I review and try to improve my system, including not to run out of Codex credits all the time, yes, all the time…

I’ve created some posts over the last month+ that I’ve just now moved to Substack, so have a look:

• When AI Makes Building Cheaper, Product and Engineering Roles Must Move
  How cheap generation creates an absorption problem: more artifacts, more review, and more pressure on product and engineering judgment.

• The New Software SOP
  Why agent-written software needs a reviewed-change loop around evidence, tests, living docs, monitoring, and learning after release.

• The New Job Is Designing the System
  How work moves from producing artifacts to designing the system that allocates context, tools, permissions, models, attention, and accountability.

• How an AI-Native Bank Gets Built
  What banking examples such as Allica, Ramp, and BBVA show about federated AI operating models, governed platforms, and production paths from local ideas to monitored applications.

• AI Cost Is an Operating Discipline
  Why AI cost cannot be cleaned up afterward by finance alone. Cost control belongs inside architecture, routing, evals, observability, and product judgment.

This is the glimpse of what I plan to continue writing on: as AI makes production easier, more of the valuable work moves into designing the system around it.

The system decides which ideas become products, which generated work is worth trusting, which workflows deserve expensive intelligence, which controls can run in software, and where humans need to make the call.

Thanks for reading!

This is a personal publication. Views expressed here are my own.