Free your mind

Cypher Hippie

[Free your mind @RobotChoice]

My agents work in Savor, the open source workspace that runs Claude Code and Codex on my machine. Checking the collection is just a rule and a small script they use before planning something new. https://t.co/0ajKCq7x0P

[Free your mind @RobotChoice]

Every model uplift makes part of my agent setup obsolete. Prompts, skills and workarounds written for the old model turn into ballast. What survives is the idea behind them. That's why a few friends and I keep a private collection of ideas, not code.

[Free your mind @RobotChoice]

From now on I'll share one of my own ideas from it every Friday. Remember - everything is an idea!

[Free your mind @RobotChoice]

Savor runs Claude Code and Codex on my machine and keeps its own memory the same way: documents, backlog and decisions are plain files in the project folder. Free and open source: https://t.co/0ajKCq7x0P

[Free your mind @RobotChoice]

Local agents (Claude Code, Codex) vs. chat apps (ChatGPT, Claude Desktop): worlds apart, above all in memory. Locally, memory is markdown files I own. I read them, fix them, version them in git, and the same rules feed both agents. In the chat apps, the AI companies own it.

[Free your mind @RobotChoice]

Savor, the open source workspace my agents work in, ships this for every code project as the Uplift check: after a model change it proposes instruction lines to drop, and brings them back when your corrections show the old mistake. https://t.co/0ajKCq7x0P

[Free your mind @RobotChoice]

Heureka! Every workaround you give an agent patches a weakness of today's model. When the next model fixes it, the patch stays and turns into ballast. So in my setup, cleaning up after a model change is a regular job, not an afterthought.

[Free your mind @RobotChoice]

Why bother: every rule costs context in every session, and more text doesn't make agents more obedient. A lean setup is cheaper, easier to follow, and I still understand it.

[Free your mind @RobotChoice]

In Savor, the open source workspace my agents work in, a scheduled workflow simply calls such a script when it has to run regularly. And once a week a workflow looks for chores I still do by hand. https://t.co/0ajKCq7x0P

[Free your mind @RobotChoice]

Agents are great at judgment and expensive at routine. New rule in my setup: a chore with fixed steps becomes a script. Agents had rebuilt the same throwaway script for TestFlight invites several times. Now one script does it the same way every time, readable and free to run.

[Free your mind @RobotChoice]

The Kontor is just one project in Savor, the open source workspace my agents work in: one conversation per task, briefs sent from project to project, workflows on a schedule. https://t.co/0ajKCq7x0P

[Free your mind @RobotChoice]

What is the Kontor in my agentic setup? It is the front door for new ideas, my strategic advisor and the assistant that sees across all my projects and hands out work when needed. It orchestrates and informs, rolls out and listens. It simply thinks along with me.

[Free your mind @RobotChoice]

Classic org development (Scrum, agile, SAFe) adds roles, ceremonies and layers so people can coordinate. Mine has two levels: I set the direction, agents act. The Kontor is no middle manager. I decide direction, money, publications and anything irreversible. The rest ships.