Stop starting over. Start compounding.
Most people use AI the same way every time. Type a question. Get an answer. Start from zero again tomorrow. The ambitious ones write better prompts. A few set preferences. Almost nobody does what actually matters: systematically teaching the AI how they think, decide, and work.
There's a reason for that. It sounds like a lot of work... and if you had that kind of time, you wouldn't need an AI in the first place.
Don't tell AI who you are. Show it how you decide.
A distillery doesn't preserve everything. It discards volume to extract essence. The AI Distillery works the same way: you feed it your raw intellectual material... sent emails, draft revisions, feedback to students, notes to yourself... and it identifies the patterns in how you actually operate. Not who you say you are. How you behave when you're working.
The output is three portable artifacts that replace the blank-slate problem with a running start. They work across tools, across models, across updates. When the AI changes, your distillation doesn't.
An operational specification for how AI should work with you. Not "be concise" but "responses over 200 words need explicit justification." Your epistemic standards, uncertainty tolerance, attribution rules, and stop conditions... all grounded in evidence.
What the AI has inferred about you, with confidence levels and receipts. Designed to be wrong in specific, correctable ways. Your corrections are the real data. The ledger makes the AI's interpretive work visible so you can fix misreadings before they calcify.
A short, reusable block you load at the start of serious work. In tools with memory, it's a context document. In tools without, you paste it. Either way, it turns a stateless system into something that knows how to work with you.
The Fermentation Engine is a prompt with instructions. Download it, paste your raw material into it, and run it in Claude, ChatGPT, or any AI you already use. You keep control. You keep your data.
When you paste work material into an AI tool, you're uploading to a third-party server. That data may be stored, used for training, or subject to breach. Know your tool's terms.
Protect yourself: Redact names before pasting ("Dear John" becomes "Dear [Student A]"). Preserve tone and structure... remove identifying details. If you wouldn't want it leaked, don't paste it.
For academics: FERPA rules may prohibit uploading student work with identifying information. IRB rules may apply if you're researching AI pedagogy. Tenure materials often contain confidential peer review. When in doubt, redact.
Current tool policies (verify before use):
AI makes stuff up. It invents citations, misremembers facts, and confabulates with confidence. In a distillation, this means the AI might infer preferences you don't have or cite "evidence" that doesn't exist.
This is by design. The engine is told to over-infer from thin evidence because your corrections are the real data. Don't be surprised when it hallucinates a preference. That's a feature.
Red flags: The AI cites specific quotes you don't recognize. All confidence levels are "High" with no uncertainty. Evidence sounds too perfectly on-theme. Your move: demand receipts. "Show me where I said that." If it can't quote the actual text, it made it up.
Don't want to paste emails? You can ask any AI to search your public work instead... published articles, professional profiles, talks, social media. It'll build the artifacts from your public persona.
Tradeoffs: No privacy risk (it's already public). But public personas are polished, and distillation works best on messy drafts. The AI can't see your corrections, feedback loops, or decision-making process. You'll get a "how I present professionally" profile, not a "how I actually work" profile.
Best for: Public intellectuals with substantial online presence, or anyone willing to accept less accuracy for zero data risk.
Using the AI Distillery in your teaching, research, or knowledge work? We're collecting dispatches from practitioners... what worked, what surprised you, what you'd do differently. The best submissions get featured here and in The Distillation Log newsletter.
No forums. No comment sections. Just curated contributions from people doing the work.
Submit Your Experience