Why Explaining a Process Is Harder Than Running It - The AI Agent New Hire Problem
Why Explaining a Process Is Harder Than Running It
Every new agent session starts from zero. It's the eternal new hire that never builds institutional memory. You spend more time explaining how to do the task than it would take to just do it yourself.
The Eternal New Hire Problem
Human employees ramp up. You explain the process once, they ask questions, make mistakes, learn, and eventually become self-sufficient. After a few weeks, they just know how things work.
AI agents never get past day one. Every session is a fresh start with no memory of previous conversations. You find yourself writing the same context, re-explaining the same conventions, and correcting the same mistakes session after session.
Why Documentation Becomes Critical
This limitation makes one thing abundantly clear - if you can't write down your process clearly enough for a zero-context agent to follow, you don't actually understand your process.
Most knowledge work runs on tribal knowledge. People know how things work because they've been around, not because it's documented. When you try to hand a task to an AI agent, all that undocumented context becomes a wall.
The fix is better documentation, but writing good process documentation is itself a skill most teams never developed because they never needed to. Now they do.
Practical Workarounds
The best approaches people have found:
- CLAUDE.md files - project-level instructions that persist across sessions and give agents baseline context
- Spec-first workflows - write detailed specifications before launching agents, instead of explaining interactively
- Template prompts - reusable prompt templates for recurring tasks that include all necessary context
- Memory systems - tools that persist key learnings between sessions so the agent isn't completely blank every time
The Deeper Lesson
The difficulty of explaining a process to an AI agent reveals how much of our work is implicit. We think we know how to do something, but when forced to make every step explicit, we realize half the process lives in our heads as intuition.
That's actually valuable. The process of trying to automate something often teaches you more about the process than years of just doing it.
Fazm is an open source macOS AI agent. Open source on GitHub.