Field notes from building with AI
HAVOC.FORK.
I don’t know how to build this.
So I’m trying it.
A notebook of real projects, wrong turns, and the work between an idea and something that actually works.
Questions. Attempts. Wrong turns.
Featured ideas
How much of my “AI agent” actually needs AI?
Before adding an API, pull the system apart. Which decisions need reasoning, and which just need a rule?
The graphic looked right. The scores were wrong.
A polished result is worthless if the facts underneath it cannot be trusted.
It started with three screenshots.
The reference images started a conversation. The useful work came after them.
Why does the video have to leave the phone?
Before designing an upload pipeline, question whether an upload is necessary.
Ideas, still in progress. Featured entries are selected from a growing notebook. Drafts and ideas remain clearly labeled until ready to publish.
The work behind the words
Project notebooks
CampSznCamp discovery & planning
01 featured idea
01How much of my “AI agent” actually needs AI?SJVarsityClubScores, schedules & systems
CampFilmVideo & computer vision
01 featured idea
04Why does the video have to leave the phone?About HavocFork
Learning in
the middle of it.
Show the work. Keep the wrong turns.
Ask a question. Try something. Record what happened.
HavocFork is about building real things with AI: researching, coding, structuring data, building agents, automating workflows, and debugging failures.
This is a record from the amateur side of the work. The experiments come from AtGameSpeed projects, where each new feature brings another question to answer.
The useful part is understanding what happened—and what to try next.