I want to consistently dedicate time to work that benefits the broader community. For me, that means contributing to open source and thinking in public. By “consistently” I mean spending a few hours on it every day.

I decided to clean up my social media workflow a bit. I’ve roughly split my posts into two categories: 1. Drafts. Raw thoughts, ideas, plans, and work in progress that I’m comfortable sharing publicly 2. Useful. Finished work, lessons learned, and conclusions that might be valuable to other people

My homepage is ridiculously simple. Just a short homepage and a blog. I use 11ty and water.css, nothing else.

All my infrastructure is described in code and deploys with one command. I use Pulumi/Terraform for cloud infrastructure and NixOS for server configuration. It used to be just convenient. Now, with AI, it’s absolutely necessary.

I’m Andy. I build infrastructure for AI agents. Making agents cheaper to run, faster to onboard, harder to kill, and better at their job. The number I’m aiming at: 8,000 agents running, each worth about $10/month to the people using them, each costing under $1/month to keep alive.

Knowledge graphs can be rebalanced, inspired by how AVL trees rebalance after insertion.

Continuous Assembly is CI/CD for the physical world.

Exclusive access to a technology cannot be maintained.

Strategy is a description of the desired state and a description of all the perspectives you need to view from.

I migrated my card system from markdown files to org-mode thread files.

Less experienced developers have no fear of hard tasks.

There are two types of writing: writer-based and reader-based.

RAG systems can have a coefficient of actuality to multiply into the cosine distance.

I'm bringing Zotero back as my reading tool.

Agents should have no local state.

I try to live as if someone is watching every second. It keeps me honest — in actions and in thoughts.

Write about everything you read, study, and think about.

Right now I open a Claude Code tab, dictate a command by voice, and while it thinks, I switch to the next tab and repeat.

I need an AI research agent that can search across Twitter, Reddit, GitHub, Product Hunt, and Hacker News.

The thinking process is more important than the result of thinking.

I believe all my ideas and notes should be public.

Goal management works like a PID controller. Measure where you are, compare to where you want to be, push in that direction. You don't need perfect precision — just continuous correction.

There are at least two ways to improve our processes with AI.

Information consumption habits matter more than people realize.

What if we could use English as a functional programming language?

What I don't know is more important than what I already know.

It's better to start simple and improve continuously than to wait until you can do it perfectly on the first try.

The only way to learn something is to start doing it.

I explore different human-AI interaction interfaces.

I'm just exploring a new way of thinking by threads.