11 posts
AI & Tools
Agent workflows and tools examined through experiments, worklogs, and practical limits.
Blog
This is one body of work organized into five editorial tracks. Each track keeps a stable voice and reading rhythm, while tags connect related ideas across the whole collection.
Subscribe to the RSS feed to stay updated with new posts.
Editorial Tracks
Every post belongs to one track with its own subject, reading rhythm, and voice. Tags connect ideas across tracks.
18 posts
Engineering Systems
Architecture, reliability, languages, and the mechanisms behind engineering choices.
16 posts
Field Notes
Personal history, retrospectives, portfolios, and observations from work in progress.
8 posts
Leadership & Teams
Ownership, autonomy, communication, and the systems that shape how teams operate.
27 posts
Quest Engine
The connected framework, from Search, Drive, and Renew to practical applications.
Far Away Is the Same as Far in the Future
You cannot drive a Mars rover with a joystick, because the round trip for a radio signal runs six to forty-four minutes. Distance and delay are one constraint measured in two different units, and once the round trip is slower than the thing you are controlling, central control stops being inefficient and becomes impossible. The physics version of the argument for autonomy at the edge.
Signal-to-Noise, Taken Literally
People praise someone for having a 100% signal-to-noise ratio. Borrowed back from the field it came from, that is not a compliment, it is an impossible measurement. What SNR actually means, why the opposite of noise is information rather than signal, and what a headset with a microphone on it says about being a transmitter and a receiver at the same time.
AI, SI, and the Problem With Renaming a Term of Art
The top labs and the current administration got together and decided to start calling it Super Intelligence instead of Artificial Intelligence, SpaceX is renaming its AI unit to SpaceXSI, and some states like California are sticking with AI. This is about the name itself rather than the politics. I actually like the SI framing, the one real argument for it is linguistic because super and intelligence travel across languages and the acronym does not flip the way IA does in Spanish, I am wary of anything that shades into compelled speech, and my only real concern is whether the benefit reaches everybody rather than the top one percent.
Goal, Effort, Verification: Three Directives on the Quest Engine Map
Boris Cherny's three prompting directives map onto the Quest Engine exactly: what you want is the Why (the problem statement), how it should verify is Renew (the definition of done), and how much effort is the How. The middle one is the subtle one, because effort is no longer a procedure, it is a strategy, a topology of agents, and a budget of resource plus time. The list is short because Search and execution are assumed.
Agents Are Taking the Drive
Roughly 80% of the software I produce now starts as a prompt, not a keystroke. Mapped onto the Quest Engine, that means agents have moved into the Drive (the How) and are eating into the four W's. What is left for me is the Why, the context still trapped in people's heads, and the work of running a fleet of agents predictably.
Elon's Algorithm: A Reference for the Manual Process
A reference write-up of Musk's five-step algorithm and what each step is supposed to mean when the thing being improved is a manual process in software engineering. Question the requirement, delete the step, simplify, accelerate, automate. Mostly a description of what the algorithm claims rather than an argument that it is correct, with a short closing note on where I do agree, in particular the rule that if you are not occasionally adding things back you are not cutting enough.
From Manual to Self-Sustaining: The Quest Engine Path
Elon's algorithm is not really an automation algorithm. It is a procedure for forcing shared understanding of a process, and automation is what it happens to permit at the end. The Quest Engine says the same thing structurally and translates it to a world that is mostly digital, where the walk around the factory floor can be delegated to agents and the goal is not a fully automated process but a self-sustaining one that keeps humans only where judgment genuinely lives.
Joy, Freedom, Motivation: Matz's Triad and the Quest Engine
In a conversation about Ruby, Rails, and the future of programming, Matz described joy, freedom, and motivation as the thing a language gives you that is hard to replicate. Freedom maps cleanly to Autonomy and Drive. Joy turns out to be the Discovery/Play/Joy triad. Motivation is not a force at all, it is the container, and it needs expanding before the mapping closes.
The Contextual Core: Who, What, When, Where
Why is the directive and How is the action, which leaves out the entire middle. Who, What, When, and Where are the locating step: four separate questions, each with its own failure mode, that tell you where you are standing before you move. Splitting them out of the How is what makes the whole model work.
The Meta Structure: How to Read the Three Steps
A high-level map of Why, the four Ws, and How, plus the two properties that make them work: the structure is self-referencing (every step contains the whole shape) and it is a loop (the end of one cycle issues the start of the next). Which also means half the vocabulary is relative to wherever you are currently standing.