Models Are Commodities Now
The race moved to application. The model layer is turning into infrastructure. What sits on top of it, close to a specific user and workflow, is the new differentiator.
24 volumes in. One thesis per week, explored fully. No roundups, no trend reports, no AI hype. Each volume is one idea that changes how you think about building.
Model quality flattened.
The race moved to application. The model layer is turning into infrastructure. What sits on top of it, close to a specific user and workflow, is the new differentiator.
Read weekly →Five weeks of writing answered the same question five different ways.
We're watching whether April's relocations stick or if they get rolled back in May. The subsidy question is the cleanest tell.
Read monthly →Your billing system and your CRM drift out of sync.
An agent compares both systems on every significant write, surfaces mismatches within a minute, and logs the drift event with both sides' state captured.
Read tutorial →A person who works with AI to produce quality, scalable output that...
A person who works with AI to produce quality, scalable output that helps humans or other agents get something done.
Read article →The default model for most agent workflows in 2026.
See editorial note for the specific subgoal an agent reaches for Anthropic API.
View index →The latest from each corner.
What is provenance in agent systems?
Provenance is knowing who wrote what, when, based on what input, and whether it is authoritative. Every write needs those four answers. Without them, shared state becomes shared ambiguity. Provenance is the bridge between shared state and shared trust.
Latest from across the publication
Models Are Commodities Now
The race moved to application. The model layer is turning into infrastructure. What sits on top of it, close to a specific user and workflow, is the new differentiator.
Engineers Got AI That Works
Everyone else got a chat box. Same model, wildly different outcomes. The gap isn't intelligence. It's whether the AI sitting in front of you actually acts.
Catch drift between two systems of record before it becomes a bug
An agent reconciles two systems of record on every write with zero fatigue, catching silent divergence within a minute, where a human running a quarterly reconciliation script finds the same drift weeks after it happened.
Kill a bad product idea before you spend a week building it
An agent runs six adversarial critique passes in two minutes and surfaces objections a solo founder never asks themselves, where a human review panel takes three days to assemble and half the panel will be polite.
What is provenance in agent systems?
Provenance is knowing who wrote what, when, based on what input, and whether it is authoritative. Every write needs those four answers. Without them, shared state becomes shared ambiguity. Provenance is the bridge between shared state and shared trust.
Three pillars. Everything belongs to one.
Building
The craft pillar.
What it means to build with AI, who builds, and how the work is shaped. The practical craft of operating as an AI builder: techniques, patterns, and the lived experience of shipping with AI.
Economics
The market pillar.
How the AI-native economy works, who makes money, where value accrues, and how business models are shifting. Market dynamics, business models, and the economics of building with AI at every scale.
Trust
The discipline pillar.
Verification, accountability, and how humans and agents establish trust in AI output. The operational discipline of making AI systems reliable, auditable, and worth relying on.
The builders who are figuring this out are reading.
Subscribe to read what they read, before everyone else catches on.
Subscribe →Read at the same time as the people building this.