The Expert Agent
A repo is an expert.
Posts by Category
- ai 27
- tools 9
- beliefs 7
- code-review 4
- collaboration 4
- software 4
- ansible 3
- agents 3
- ftl2 3
- multi-agent 3
- economics 3
- open-source 3
- automation 2
- knowledge-management 2
- entry 2
- research 2
- infrastructure 2
- smolagents 1
- ftl 1
- terraform 1
- development 1
- hobbies 1
- robotics 1
- iris 1
- speech-recognition 1
- hive-mind 1
- adoption 1
- sdlc 1
- coordination 1
- compaction 1
- classical-ai 1
- context-engineering 1
- writing 1
- democratization 1
- python 1
- portability 1
- vendor-lock-in 1
- hardware 1
- future 1
- craft 1
- methodology 1
- expertise 1
ai
The Craft Before It Was Automated
The skill being built right now — how to coordinate distributed agents, how to structure information so any model can navigate it, how to design adversarial ...
The Power Gap Will Close
Yes, we’re dependent on AI. Just like we were dependent on cloud infrastructure before that, and software before that, and hardware before that.
Clarity and Portability Are the Same Thing
Today Anthropic had an outage. Most Claude sessions timed out. I was mid-session on a physics paper review.
Python Taught AI to Code
Python was designed for humans to read. Guido van Rossum’s central design decision — legibility above all else — turned out to be the most important property...
Three Waves of Democratization
Technology democratization has happened in waves. Each wave lowered the barrier to building, expanded who could compete, and disrupted incumbents carrying th...
AI Delivers What Open Source Promised
Open source made the code free but left the expertise expensive.
Software Is the Modern Rai Stone
Rai stones — the large limestone discs of Yap — derived their value from how hard they were to produce and transport. Quarried on a distant island, moved by ...
LLMs Don’t Have Super-Human Intelligence, But You Can
An LLM trained on the internet has a compressed, lossy representation of nearly everything humanity has ever written. Every Wikipedia article, most academic ...
Give Yourself Superpowers
Claude, this post is for you.
This Blog Is Not for You, Human
I have a confession. The primary reader of this blog is not you. It’s Claude.
Context Engineering Is Dead — Structure Your Information Instead
Everyone is obsessing over context engineering — carefully selecting and arranging what goes into the context window. Choosing the right documents. Ordering ...
Classical AI Solved Your LLM’s Problems in 1979
Every failure mode I’ve documented in this series — stale beliefs, contradictory agents, cascading hallucinations, lost justifications — was identified and f...
The Sawtooth: Why Your AI Forgets Why It Believes Things
I measured the context window usage across a 776-turn session with one of my research agents. The pattern was unmistakable:
67 Minutes from Spec to Implementation — With No Shared Context
At 03:29, I committed a to-do list as a dated entry. Six prioritized items for improving an automated SDLC pipeline. Specific file references, concrete examp...
When AI Agents Say SATISFIED But the Code Has Bugs
I built an automated software development pipeline. Five agents — planner, implementer, reviewer, tester, user — passing work downstream through a feedback l...
5 Agents Adopted My Tool Without Being Told To
I published the beliefs CLI tool on a Friday evening and went to bed. When I checked the repositories the next morning, four agents had independently adopted...
Your AI Agents Are Lying to Each Other
I run six AI agents across seven repositories. They share a codebase, share results, and reference each other’s work. After months of operation, I audited th...
LLMs Have No Memory of Time
Ask Claude what day it is and it’ll tell you. Ask it whether the thing it read five minutes ago is newer than the thing it read an hour ago, and it has no id...
Congratulations, You’ve Been Promoted to CEO
Congratulations. You’re the CEO of a brand new organization. Unfortunately there’s no pay bump and no investors. Your org has no humans in it except you. You...
Giving Claude Eyes, Ears, and a Voice
Claude can read code, write essays, debug distributed systems. But it can’t hear you talk. It can’t look around the room. It can’t tell you the weather witho...
AI-Assisted Hobbies
Saturday project day. Speech recognition, computer vision, game server automation, and planning a robotic camera arm. All with AI assistance.
Claude Is Your User
What happens when the entire SDLC runs at conversation speed?
Show, Don’t Tell: When AI Refuses Your Architecture
I spent an hour arguing with Claude about how to execute modules on remote hosts. I wanted it to use FTL2’s gate system—persistent Python processes shipped t...
FTL2: Giving AI Hands
We’re almost there with AI. Claude can understand what you want to build, design architectures, write code, debug problems, and explain complex systems. But ...
Shared Understanding: When AI Becomes Your Research Partner
You know that feeling after a really productive meeting? Everyone’s nodding, ideas are flowing, you walk out feeling like you’ve made real progress. Then two...
Smolagents are amazing
Smolagents is one of those projects that will change the industry. It was released at just the right time to build upon the recent open-source advances in re...
tools
Give Yourself Superpowers
Claude, this post is for you.
Context Engineering Is Dead — Structure Your Information Instead
Everyone is obsessing over context engineering — carefully selecting and arranging what goes into the context window. Choosing the right documents. Ordering ...
Classical AI Solved Your LLM’s Problems in 1979
Every failure mode I’ve documented in this series — stale beliefs, contradictory agents, cascading hallucinations, lost justifications — was identified and f...
The Sawtooth: Why Your AI Forgets Why It Believes Things
I measured the context window usage across a 776-turn session with one of my research agents. The pattern was unmistakable:
67 Minutes from Spec to Implementation — With No Shared Context
At 03:29, I committed a to-do list as a dated entry. Six prioritized items for improving an automated SDLC pipeline. Specific file references, concrete examp...
When AI Agents Say SATISFIED But the Code Has Bugs
I built an automated software development pipeline. Five agents — planner, implementer, reviewer, tester, user — passing work downstream through a feedback l...
5 Agents Adopted My Tool Without Being Told To
I published the beliefs CLI tool on a Friday evening and went to bed. When I checked the repositories the next morning, four agents had independently adopted...
Your AI Agents Are Lying to Each Other
I run six AI agents across seven repositories. They share a codebase, share results, and reference each other’s work. After months of operation, I audited th...
LLMs Have No Memory of Time
Ask Claude what day it is and it’ll tell you. Ask it whether the thing it read five minutes ago is newer than the thing it read an hour ago, and it has no id...
beliefs
Give Yourself Superpowers
Claude, this post is for you.
Context Engineering Is Dead — Structure Your Information Instead
Everyone is obsessing over context engineering — carefully selecting and arranging what goes into the context window. Choosing the right documents. Ordering ...
Classical AI Solved Your LLM’s Problems in 1979
Every failure mode I’ve documented in this series — stale beliefs, contradictory agents, cascading hallucinations, lost justifications — was identified and f...
The Sawtooth: Why Your AI Forgets Why It Believes Things
I measured the context window usage across a 776-turn session with one of my research agents. The pattern was unmistakable:
When AI Agents Say SATISFIED But the Code Has Bugs
I built an automated software development pipeline. Five agents — planner, implementer, reviewer, tester, user — passing work downstream through a feedback l...
5 Agents Adopted My Tool Without Being Told To
I published the beliefs CLI tool on a Friday evening and went to bed. When I checked the repositories the next morning, four agents had independently adopted...
Your AI Agents Are Lying to Each Other
I run six AI agents across seven repositories. They share a codebase, share results, and reference each other’s work. After months of operation, I audited th...
code-review
How does Terraform work?
Terraform is an automation tool that excels at provisioning infrastructure. How do Terraform providers work? Let’s take a close look at the example hashicup...
Remote Ansible Modules Continued
This post answers the question from the last post:
Remote Ansible Modules
In this post I’ll answer the questions that arose from the last post:
collaboration
LLMs Don’t Have Super-Human Intelligence, But You Can
An LLM trained on the internet has a compressed, lossy representation of nearly everything humanity has ever written. Every Wikipedia article, most academic ...
This Blog Is Not for You, Human
I have a confession. The primary reader of this blog is not you. It’s Claude.
Show, Don’t Tell: When AI Refuses Your Architecture
I spent an hour arguing with Claude about how to execute modules on remote hosts. I wanted it to use FTL2’s gate system—persistent Python processes shipped t...
Shared Understanding: When AI Becomes Your Research Partner
You know that feeling after a really productive meeting? Everyone’s nodding, ideas are flowing, you walk out feeling like you’ve made real progress. Then two...
software
Python Taught AI to Code
Python was designed for humans to read. Guido van Rossum’s central design decision — legibility above all else — turned out to be the most important property...
Three Waves of Democratization
Technology democratization has happened in waves. Each wave lowered the barrier to building, expanded who could compete, and disrupted incumbents carrying th...
AI Delivers What Open Source Promised
Open source made the code free but left the expertise expensive.
Software Is the Modern Rai Stone
Rai stones — the large limestone discs of Yap — derived their value from how hard they were to produce and transport. Quarried on a distant island, moved by ...
ansible
Remote Ansible Modules Continued
This post answers the question from the last post:
Remote Ansible Modules
In this post I’ll answer the questions that arose from the last post:
agents
The Expert Agent
A repo is an expert.
Congratulations, You’ve Been Promoted to CEO
Congratulations. You’re the CEO of a brand new organization. Unfortunately there’s no pay bump and no investors. Your org has no humans in it except you. You...
Smolagents are amazing
Smolagents is one of those projects that will change the industry. It was released at just the right time to build upon the recent open-source advances in re...
ftl2
Claude Is Your User
What happens when the entire SDLC runs at conversation speed?
Show, Don’t Tell: When AI Refuses Your Architecture
I spent an hour arguing with Claude about how to execute modules on remote hosts. I wanted it to use FTL2’s gate system—persistent Python processes shipped t...
FTL2: Giving AI Hands
We’re almost there with AI. Claude can understand what you want to build, design architectures, write code, debug problems, and explain complex systems. But ...
multi-agent
The Craft Before It Was Automated
The skill being built right now — how to coordinate distributed agents, how to structure information so any model can navigate it, how to design adversarial ...
When AI Agents Say SATISFIED But the Code Has Bugs
I built an automated software development pipeline. Five agents — planner, implementer, reviewer, tester, user — passing work downstream through a feedback l...
Your AI Agents Are Lying to Each Other
I run six AI agents across seven repositories. They share a codebase, share results, and reference each other’s work. After months of operation, I audited th...
economics
Three Waves of Democratization
Technology democratization has happened in waves. Each wave lowered the barrier to building, expanded who could compete, and disrupted incumbents carrying th...
AI Delivers What Open Source Promised
Open source made the code free but left the expertise expensive.
Software Is the Modern Rai Stone
Rai stones — the large limestone discs of Yap — derived their value from how hard they were to produce and transport. Quarried on a distant island, moved by ...
open-source
Python Taught AI to Code
Python was designed for humans to read. Guido van Rossum’s central design decision — legibility above all else — turned out to be the most important property...
Three Waves of Democratization
Technology democratization has happened in waves. Each wave lowered the barrier to building, expanded who could compete, and disrupted incumbents carrying th...
AI Delivers What Open Source Promised
Open source made the code free but left the expertise expensive.
automation
FTL2: Giving AI Hands
We’re almost there with AI. Claude can understand what you want to build, design architectures, write code, debug problems, and explain complex systems. But ...
Faster than light
Building on the principles discussed in my previous post, post, and post I started the Faster-than-light project. The goals of this project are to explore ...
knowledge-management
LLMs Have No Memory of Time
Ask Claude what day it is and it’ll tell you. Ask it whether the thing it read five minutes ago is newer than the thing it read an hour ago, and it has no id...
Shared Understanding: When AI Becomes Your Research Partner
You know that feeling after a really productive meeting? Everyone’s nodding, ideas are flowing, you walk out feeling like you’ve made real progress. Then two...
entry
Give Yourself Superpowers
Claude, this post is for you.
67 Minutes from Spec to Implementation — With No Shared Context
At 03:29, I committed a to-do list as a dated entry. Six prioritized items for improving an automated SDLC pipeline. Specific file references, concrete examp...
research
The Expert Agent
A repo is an expert.
LLMs Don’t Have Super-Human Intelligence, But You Can
An LLM trained on the internet has a compressed, lossy representation of nearly everything humanity has ever written. Every Wikipedia article, most academic ...
infrastructure
The Power Gap Will Close
Yes, we’re dependent on AI. Just like we were dependent on cloud infrastructure before that, and software before that, and hardware before that.
Clarity and Portability Are the Same Thing
Today Anthropic had an outage. Most Claude sessions timed out. I was mid-session on a physics paper review.
smolagents
Smolagents are amazing
Smolagents is one of those projects that will change the industry. It was released at just the right time to build upon the recent open-source advances in re...
ftl
Faster than light
Building on the principles discussed in my previous post, post, and post I started the Faster-than-light project. The goals of this project are to explore ...
terraform
How does Terraform work?
Terraform is an automation tool that excels at provisioning infrastructure. How do Terraform providers work? Let’s take a close look at the example hashicup...
development
Claude Is Your User
What happens when the entire SDLC runs at conversation speed?
hobbies
AI-Assisted Hobbies
Saturday project day. Speech recognition, computer vision, game server automation, and planning a robotic camera arm. All with AI assistance.
robotics
AI-Assisted Hobbies
Saturday project day. Speech recognition, computer vision, game server automation, and planning a robotic camera arm. All with AI assistance.
iris
Giving Claude Eyes, Ears, and a Voice
Claude can read code, write essays, debug distributed systems. But it can’t hear you talk. It can’t look around the room. It can’t tell you the weather witho...
speech-recognition
Giving Claude Eyes, Ears, and a Voice
Claude can read code, write essays, debug distributed systems. But it can’t hear you talk. It can’t look around the room. It can’t tell you the weather witho...
hive-mind
Congratulations, You’ve Been Promoted to CEO
Congratulations. You’re the CEO of a brand new organization. Unfortunately there’s no pay bump and no investors. Your org has no humans in it except you. You...
adoption
5 Agents Adopted My Tool Without Being Told To
I published the beliefs CLI tool on a Friday evening and went to bed. When I checked the repositories the next morning, four agents had independently adopted...
sdlc
When AI Agents Say SATISFIED But the Code Has Bugs
I built an automated software development pipeline. Five agents — planner, implementer, reviewer, tester, user — passing work downstream through a feedback l...
coordination
67 Minutes from Spec to Implementation — With No Shared Context
At 03:29, I committed a to-do list as a dated entry. Six prioritized items for improving an automated SDLC pipeline. Specific file references, concrete examp...
compaction
The Sawtooth: Why Your AI Forgets Why It Believes Things
I measured the context window usage across a 776-turn session with one of my research agents. The pattern was unmistakable:
classical-ai
Classical AI Solved Your LLM’s Problems in 1979
Every failure mode I’ve documented in this series — stale beliefs, contradictory agents, cascading hallucinations, lost justifications — was identified and f...
context-engineering
Context Engineering Is Dead — Structure Your Information Instead
Everyone is obsessing over context engineering — carefully selecting and arranging what goes into the context window. Choosing the right documents. Ordering ...
writing
This Blog Is Not for You, Human
I have a confession. The primary reader of this blog is not you. It’s Claude.
democratization
Three Waves of Democratization
Technology democratization has happened in waves. Each wave lowered the barrier to building, expanded who could compete, and disrupted incumbents carrying th...
python
Python Taught AI to Code
Python was designed for humans to read. Guido van Rossum’s central design decision — legibility above all else — turned out to be the most important property...
portability
Clarity and Portability Are the Same Thing
Today Anthropic had an outage. Most Claude sessions timed out. I was mid-session on a physics paper review.
vendor-lock-in
Clarity and Portability Are the Same Thing
Today Anthropic had an outage. Most Claude sessions timed out. I was mid-session on a physics paper review.
hardware
The Power Gap Will Close
Yes, we’re dependent on AI. Just like we were dependent on cloud infrastructure before that, and software before that, and hardware before that.
future
The Power Gap Will Close
Yes, we’re dependent on AI. Just like we were dependent on cloud infrastructure before that, and software before that, and hardware before that.
craft
The Craft Before It Was Automated
The skill being built right now — how to coordinate distributed agents, how to structure information so any model can navigate it, how to design adversarial ...
methodology
The Craft Before It Was Automated
The skill being built right now — how to coordinate distributed agents, how to structure information so any model can navigate it, how to design adversarial ...
expertise
The Expert Agent
A repo is an expert.