Tips

Techniques

AIs don’t think like you do

In fact, they don't think at all As useful—and oftentimes seemingly intuitive—as LLM interaction is, it is not understanding you.

Prompt Chaining as Analytical Scaffolding

How to build deep analysis one controlled layer at a time We’re used to thinking of analysis as something linear: you pose a question, crunch some numbers, and get an answer. But when you’re working with AI, that model breaks.

Surface assumptions

Make assumptions explicit, and then refine them AI models don’t “think” in the human sense. But they do embed assumptions—lots and lots of them. Hidden priors, structural defaults, causal heuristics, statistical norms. Often, when you ask the AI a question and it responds with confidence, what it’s really doing is applying a stack of unspoken assumptions to your input.

Shift between perspectives

What you see depends on where you're looking from Remember the old fable of the blind men and the elephant? There are ways of looking at a thing that can somehow miss its essential nature. The best protection against this is looking from multiple angles.

Push for confidence assessments

An assertion without an assessment is just an opinion An AI will always try to answer your question (as it understands it). When it replies, its answer "feels" confident: a flat, bold assertion that yes, this is true, or option B is "likely." But how useful are these assertions? Hardly at all.

Create a debate

Open your analysis to different points of view Persona creation and invocation is a next-level AI skill. If used thoughtfully, it will propel your analysis toward larger, deeper and more thoughtful conclusions.

Don’t trust. Verify.

AI is a brilliant and unreliable intellectual partner AI is smarter than anyone you've ever met. And dumber.

A Tiny Translucent Portal

Long AI conversations often end up going weird and wonky. Here's some strategies you can use.

What is “context” for an AI?

It's not what you probably think it is The AI doesn't just consult your context—it thinks with it.

Tips

Set your tone

AIs will interact with you in the way you find comfortable and productive. Tell it what you like The thing to remember about AIs is that they are general tools. Some people use them for web search, some for advice, and others to cheat on their homework.

Preserving clean context space

Without care, context can become muddy. Give the AI some rules. All of the AI systems with which I work allow forking conversational threads. I have formalized this, a bit, by declaring side-threads which may have one or more of the following three attributes:

Standardize your nomenclature

Avoiding the dreaded "semantic drift" AI systems exhibit a peculiar talent for latching onto ambiguous terms and riding them into interpretive chaos. Like a dog with a favorite stick, once an AI fixates on a particular reading of a word, it will fetch that interpretation relentlessly, regardless of shifting context.

Create a protocol document

Building a custom working environment that actually works for you Create a session protocol that turns your AI into the perfect collaborator for your specific task and work style.

Avoid trigger words

Certain words can derail a conversation: a guide Certain words are a trap that can put conversations into a death spiral AI systems carry invisible tripwires—innocuous words that instantly derail productive analysis by activating rigid response templates. Like accidentally hitting a car alarm, these trigger words transform your focused collaborator into a verbose, formulaic automaton following scripts you never requested.

“Shepherding” as a collaborative style

Deep analysis requires more than just analysis Real analytical thinking emerges not from isolated processing, but from the patient cultivation of understanding through guided conversation.