Principles

Small ideas with big impact. These should apply wherever my work is found.

• 01 Use code for everything you can Use deterministic tools for deterministic work, and spend model calls where language, judgment, or synthesis actually matter.
• 02 Every LLM session will build towards the next Each agent session should leave behind structure, notes, code, or context that makes the next session cheaper and better.
• 03 Directives are not enforced by prompting If an instruction about model output or actions can be checked mechanically, check it mechanically. Prompts are only suggestions, and very loose ones at that.
• 04 Import others' ideas, not their implementations Automations have little value if not undersood. Instead of copying someone else's automation wholesale, destill the main points and iteratively grow it to your needs from there.
• 05 Agents must have the same context as Humans The goal is for Agents and Humans to be able to work side by side, both staying in sync as the context of the business evolves. When one discovers something, it updates the other.