NIGEL ANDRÉ / ONE YEAR / BUILT IN PUBLIC
I build things, and break them, to understand how they really work.
I’m a finance and operations executive—CFO, Treasurer, and GM. My experience spans three industries—fleet leasing, distribution, and payments.
I spent a year putting AI to work across support, engineering, and reporting—different models, different applications—and building and stressing the infrastructure and networks underneath it to find where the boundaries hold.
This site is the record: the projects are the evidence, and the decisions, failures, and recoveries are what they taught.
Latest insights
-
AI governance
The Quiet Governor
Unix, Linux, AI, and the Infrastructure of Interpretation
When did we decide a machine feeling authoritative was only good news?
-
Judgment
Great minds think alike. My grandfather reminded me of another saying.
Two people reaching the same conclusion doesn’t make either of them right.
-
Operations
The Benefits of Getting It Right
Some of the most valuable information lies in the exceptions.
Mental models
The ideas that survived contact with reality.
The projects are evidence. The learning is the product.
The site is organized around decisions rather than technologies: what problem existed, what was known, what was assumed, where the boundary failed, and what became reusable afterward.
What that produces for a business is judgment in five places:
- what AI and LLM systems can safely be trusted to do
- engineered controls that sit in the operating path
- audit transparency that survives inspection
- stronger work across organizational and technical seams
- a complete operating-model view of how decisions move through a business
SYSTEM MAP
See how the pieces connect.
The projects are connected by three recurring questions: where something runs, who may act, and what information moves.
The year in review
From isolated experiments to a connected operating system.
Featured work
Systems that forced better thinking.
The tangible production and the writing tell the same story: 60K+ lines of maintained code and 13K+ lines of markup and configuration sit alongside a 12-part LinkedIn series and epilogue and The Quiet Governor.
Identity and access
Bridging incompatible trust domains, preserving attribution into Linux, and supporting application entitlements.
Network
Segmentation, separate power paths, and naming conventions that make connectivity intentional and explainable.
Compute and operations
A three-node HA cluster supporting 10 VMs across six VLANs, with redundant DNS, VPN access, UPS resilience, and recovery runbooks.
HomeLabRemote
Select Linux hosts and run approved maintenance actions at fleet scale, while each host retains final authority.
iMessage Locker
Draft replies with a local model while preserving the trust contract people already have with Messages.
Operator Confirmation
A simulation of how authority can transfer through reasonable decisions.
Retro Arcade Hub
A shared entrance for separate games, with an explicit handoff from choosing a cabinet to playing it.
The point was never to know everything.
It was to become better at finding the boundary, testing the assumption, and building a path back when the first answer was wrong.
If that way of working is useful to you, the background is on the About page and I am glad to talk.
I let AI name the product suite. In retrospect, Blame Machine should have been a warning.