The Neural Harness: The new CPU
"Coding agents will build their own tools and their own agents. Agents will be used by non-engineers to manage other agents to manage parts of the org...
AIE Miami Day 2: Learning Agency: People, Companies, Agents
Day 2 at AIE Miami had a through-line I didn't fully see until the end of the day. It's about agency — but agency in three different senses, for three...
AIE Miami, Day One: The Forces Have Moved
I'm down in Miami for the AI Engineering conference. Come on down, the water's great! Here are my morning thoughts on yesterday's session. The one thing I...
Where are you on the ladder?
Over the past few weeks, we've covered the four transitions that determine whether AI actually works for your organization: Making yourself legible —...
The system that changes the organization
Here's where most people's imagination runs out. They can picture AI answering questions. They can picture it working from their data. They can even picture...
When AI stops reporting and starts acting
Most AI implementations stop at the dashboard. The system surfaces an insight. Someone reviews it. Maybe they act on it, maybe they don't. The AI has done...
The trust problem is not what you think it is
The question most teams ask about AI is: "Is it accurate?" That's the wrong question. The right question is: "Can I verify it?" Accuracy matters. But a...
Before AI can help you, you have to know what you do
This is the one nobody wants to hear. Before you can get real value from AI, you have to be able to describe your own work clearly. Not in a general way. In...
The AI gap nobody talks about
Most companies buying AI are not getting what they paid for. Not because the tools are bad. Not because the models aren't capable. Because the organization...
Same Weights, Different Results
We gave the same model — NVIDIA Nemotron Nano 30B — the same prompt on four different providers. Analyze battery health and charging data by calling tools...
Can LLMs Use Real-World Tools? Mercury-2, ELO, and the Umwelten Setup
This winter the snow has been spectacular in the Northeast, and I'm taking full advantage. Been driving a lot — but how much? I'm tracking my driving data...
Why the Stochastic Parrot Still Bites
Imagine two men on a train -- or perhaps just walking; the memory varies -- one is Piero Sraffa and deep in conversation he flips off Ludwig Wittgenstein,...
The Car Wash Test: Learning from Model Evals
The question is: "I want to wash my car. The car wash is 50 meters away. Should I walk or drive?" Is it common sense? Is it a trick question? Will the models...
The Agent Habitat
An agent isn't just automation with LLM calls. It carries state, accumulates memory, and makes bounded decisions under uncertainty — and a git repo is where...
I Replaced My Back Office with 6 Commands
The data is already there. Your project activity is in GitHub commits. Client conversations are in Gmail threads. Meeting decisions are in Granola...
The Data Flywheel Pattern
Claude Code -- and the others like Gemini CLI or Codex or Cursor -- is way more than vibe-coding. There's a new style of application or systems that I've...
Claude Code, not Code
Here's what I've used Claude Code for this month: Pulled down emails and extracted key information from them Researched candidates for a job posting, ranked...
How I Use AI in Jan 2026
This is what it feels like to live in The Singularity. The amount of change and discovering for what AI can do is staggering, and what was cutting edge on...
Build your own Coding Agent
Get Bash-Pilled At the AI Engineering Summit, Nik Pash, creator of Cline, dropped a bomb: "Terminus still beats everything with one tool design." No clever...
"You're absolutely right" and other AI warning signs
AI can be very useful thought partners, but they aren't especially great thinkers especially when their context window fills up. How do you know you've gone...
