Writing

The long versions live on Jeff’s site. This is the map.

Jeff writes at length at jeffpinto.com/notes. We do not copy those notes here, because two sites carrying the same words is worse than one site carrying them well. This page is a curated route in, organised by what you might be trying to solve.

Almost all of it is written the same way: what we did, what happened, and what we got wrong. The last part is the reason any of it is worth reading.

Measurement and AI

How you get a number out of a system that will not simply tell you one, and how you defend it afterwards.

Is your AI as trustworthy as the day it launched? The trustworthiness health card: which kind of trust moved, for which users, with arithmetic anyone can check and an interval that admits what it does not know. Why a transparency score cannot tell you what is in the corpus A disclosure index measures a developer’s paperwork, not their data. What measurement looks like instead. The audit that ships The runnable audit beats the rigorous audit that never runs. Friction is the variable. Metric drift belongs on an on-call rotation Treat model metrics like infrastructure and what a team is willing to ship changes. Privacy by default What privacy-first actually costs and what it buys, worked through on composite examples. Decoupled classifiers Closing a fairness gap without hand-waving, and the 140 permutations behind one defensible answer. The auditable recommender A recommender that can explain any single decision, built under the first Blue Camel Consulting. Procurement is asking for this again.

Education and enrollment

Institutions spending money on incomplete information, and the vocabulary problem underneath it.

The boast and the bound The 2012 Grand Canyon University pilot. Four channels, one quarter, and a word invented for a number everyone was already negotiating. Enrollment funnels Where the inquiry-to-start funnel actually leaks, and which leaks are worth fixing. Be Billy Beane Republished from a 2011 post. Marketing analytics before anyone called it that. Validation rules earn their keep An optimiser is only as good as its inputs, and the inputs are usually worse than anyone admits.

Operations and delivery

The work that pays for the slide, and does not appear on it.

The work nobody puts on a slide A third-party logistics turnaround is two years of small reconciliations. Skulogix, 2000 to 2004. Eight analysts and a courier-elasticity model A marketplace launch at Uber that looked like a growth story and was a supply problem. Rules engines When the business logic belongs in a table a human can edit on a Thursday. A 2009 procurement plan, re-read as a 2026 checklist The Ontario Medical Association work, and the four questions modern AI procurement keeps skipping. Clients buy out of fear or greed, but mostly fear An old IBM lesson, the loss-aversion math behind it, and the greed lever found inside a privacy audit pitch.

Ownership in the AI age

The argument this practice is partly a test of. Sixteen notes, indexed below.

The Owner’s Toolkit for the AI Age The index and reading guide. Own the means of production, not a check. Everyone has one opus The cure Jeff actually believes, and why the technology finally makes it possible. The commons need not be a tragedy Ostrom’s eight principles, and who owns the layer one person cannot build. A prompt is a character sheet What an agent actually is, argued through a working toy.

Contact

If one of these landed, write to jeff@bluecamelconsulting.com. Disagreement welcome.