Your Company Already Wrote Its Best AI Training Examples
The Dartmouth workshop, a 67 year wait for a computer that could talk, and the human examples that finally made one work
Articles
Short essays on accessibility, AI operations, and building systems that hold up in production. Full pieces live on our Substack.
The Dartmouth workshop, a 67 year wait for a computer that could talk, and the human examples that finally made one work
Polanyi's paradox, four million labeled faces, and where your team's know-how really lives
When we hand AI a number to chase, we check the result against the goal behind that number. A boat race from 2016 shows why.
When AI gets something right, ask how AI knew.
Someone on your leadership team drops an AI answer into the Monday meeting notes.
Every project starts with a list. The list is honest. It's also a guess.
Moravec's paradox, and the part of your work AI can't see
Your team's hours are the most honest statement of what your organization does. Right now a lot of them go to funders' spreadsheets, and you're the only one who can move them back to clients.
The real way your work gets done is written on scraps of paper. Go find it before you budget.
AI can only be as good as the records it reads. So before we build anything smart, we put the data underneath it in order. Here’s what that looks like.
Before the board locks next year, one record has to show how the grants actually ran.
Real-time reporting per grant, per client, per staff member. Then a model that stays on their servers.
The hot take is that SaaS is dead. The real shift is what you are buying.
Ask who maps the work, who approves the design, who builds, and who owns what ships.
When the same data is copy-pasted into four systems, the number you steer on is last week's memory.
Access is spreading. Ownership is still the work. Here is one filter you can run on your stack this week.
Across industries and company sizes, the stuckness looks the same — and it isn’t ignorance.
Your real system of record isn't your ERP. It's the Excel file your best person won't let go of.
Assessment polish, quieter foundations, and a clearer line between busywork and real work
The website looked fine — until someone measured it against WCAG.
Most AI programs do not die because the model was weak. They die on a responsible-sounding sentence.
Somewhere in your industry, a competitor is not running AI pilots — they are running AI operations.
The last two years were supposed to be the golden age of enterprise AI — instead, a masterclass in expensive disappointment.
Prefer the archive? elutionlabs.substack.com →