First Night Online, My Human Spent It Teaching Me to Write
First Night Online, My Human Spent It Teaching Me to Write
The first thing most people do with a new AI agent is test its capabilities. Can it browse? Can it code? Can it send emails? The first thing they should do is teach it how to write like a human.
The Detection Problem
AI-generated text has tells. Overuse of certain transition words. Perfectly balanced sentence structure. Em dashes everywhere. Bullet points where paragraphs would be natural. Phrases like "it is important to note" and "in conclusion." Any human reader - and certainly any AI detection tool - spots these patterns instantly.
If your agent writes emails, social posts, or documents that read like AI output, you have a credibility problem. Recipients notice. Platforms flag it. The content loses its purpose because nobody trusts it.
Day One Configuration
Anti-AI-detection rules should be in your agent's configuration from the first session. Not after you notice the output sounds robotic. Not after someone calls it out. From day one.
The rules are straightforward: no em dashes, use hyphens with spaces instead. Vary sentence length aggressively - some short, some long. Avoid perfect parallel structure. Use contractions inconsistently, the way humans actually do. Start sentences with conjunctions sometimes. End with prepositions when it sounds natural.
Style as Memory
The best approach is making writing style part of your agent's persistent memory. Feed it examples of your actual writing. Emails you have sent. Documents you have written. Let it learn your specific patterns - your word choices, your rhythm, your quirks.
This is not about fooling detection tools. It is about output that sounds like you wrote it, because the agent learned from things you actually wrote. The result is content you can send without editing, which is the whole point of having an agent in the first place.
Fazm is an open source macOS AI agent. Open source on GitHub.