Ai Creativity February 20, 2026

The AI-PKM Stack: Using AI as Your Junior Researcher

The framing matters here, so let me be precise about it from the beginning. I use AI as a junior researcher. Not as a writer. Not as a thinker. Not as a replacement for any part of the intellectual work that is actually mine to do. As a junior researcher, someone who handles the preparatory and organizational tasks that slow me down, so that I can spend my time on the tasks that require my specific experience, judgment, and perspective.

This framing changes how you interact with AI tools significantly. If you approach them as writers, you end up with content that sounds generic, because generic is the best an average of all writing can produce. If you approach them as junior researchers, you end up with faster access to your own thinking, which is what they are actually good at producing.

What a Junior Researcher Does

A good junior researcher does not generate original insight. They find relevant material, organize it for easy review, prepare summaries of complex sources, identify connections across multiple documents, draft outlines that give you something to react to, and handle the mechanical tasks of research so that the senior researcher can focus on interpretation and argument.

This is exactly what AI does well. It can summarize a long document in the time it would take you to read the first three pages. It can identify common themes across multiple sources you paste in. It can draft a first outline from bullet points you provide, giving you something to push against rather than a blank page to start from. It can generate counterarguments to a position you have taken, so you can stress-test it before publishing. It can clean up a voice memo transcript into something readable.

None of these tasks require the AI to have original ideas. They require the AI to handle material you provide and return it in a more useful form. This is the legitimate and powerful use of AI in a knowledge work context.

My Specific Workflow

I use Claude for most of my AI-assisted knowledge work, primarily because it handles long-form text well and its reasoning about complex ideas is useful for my specific purposes.

My typical workflow: I capture in Obsidian as normal. When I am ready to develop a thinking note into something more complete, I paste the note into Claude along with two or three related notes and ask it to identify the argument I am making, any gaps in the reasoning, and three counterarguments worth addressing. It returns a structured analysis. I respond to that analysis in my own voice, adding, removing, and revising based on my judgment. The result is my thinking, made sharper by having been stress-tested.

I also use AI for distillation support: I paste a long note or set of notes and ask for a summary of the main ideas. This gives me a starting point for my own summary, which I then rewrite in my own words. The AI’s version is rarely quite right, it misses the things that are interesting to me specifically, but it is faster than starting from nothing, and the gaps between what it notices and what I notice are often where the most interesting thinking happens.

The Test for Whether You Are Using AI Well

Ask yourself: did I bring something specific to this conversation? A claim I have been developing? An observation I have made through my own experience? A question I have been carrying for weeks?

If yes, AI makes you sharper. It has specific material to work with and can return something genuinely useful.

If you arrived with a blank page and asked AI to fill it, you got generic output because you gave generic input. Specificity is the input. Leverage is the output.

What AI Cannot Do in Your PKM

AI cannot tell you what is interesting to you. It can identify what is common, what is widely discussed, what most people in a given domain believe, but it has no access to your specific context, your particular experience, the specific angle that makes a piece of work uniquely yours. Those are the inputs AI requires from you, not outputs it can generate for you.

AI also cannot maintain continuity across extended time the way a knowledge system can. It has no memory between sessions. It does not know what you believed last month or how your thinking has developed. It cannot surface a connection between a note from two years ago and your current project. Your PKM system does all of these things. AI enhances the processing of what the system contains; it cannot replace the system itself.

Use AI to go faster. Use your knowledge system to go deeper. The combination is genuinely powerful, but only if you understand which tool does which job.

Where to Go Next

Continue with the most relevant essays, key site pages, and trusted references.

Core Pages

Trusted Sources

The Weekly Field Note

Ideas worth thinking about — in your inbox.

No noise. One insight per week on knowledge systems, creative output, and building with clarity. Join The Field Note

Part of the system — Stage III · Activate

Pillar 07: AI Collaboration

How do I use AI in my PKM?

Next in this pillar: AI Won't Replace You. But It Will Expose You. → View the full system map