Personal Knowledge Management in the LLM Era

Redesigning PKM Systems for Human-AI Cognitive Partnership

The PKM landscape is bifurcating sharply. Most tools are simply bolting AI features onto traditional frameworks—adding semantic search and chat interfaces without rethinking core assumptions. Meanwhile, a small vanguard of practitioners argues the entire paradigm needs reinvention. The “Ground → Capture → Think → Recall” workflow represents a genuinely novel contribution to this discourse, particularly in its front-loading of research and its treatment of AI conversation as a primary thinking medium rather than a retrieval tool.

The central tension in AI-era PKM is this: the activities that make traditional PKM valuable—manual linking, deliberate organization, effortful synthesis—are precisely what AI automates away. Research shows a -0.68 correlation between AI tool usage and critical thinking abilities. Yet those who ignore AI’s capabilities will be outpaced by those who leverage them. The solution isn’t choosing between human effort and AI assistance—it’s designing workflows where AI handles information retrieval and pattern recognition while humans maintain ownership of synthesis, meaning-making, and epistemic judgment.

The Current Landscape: A Three-Tier Spectrum of AI Integration

The PKM tool ecosystem in early 2026 falls into three distinct categories. At the traditional end, tools like plain-text Obsidian and original GTD apps operate on manual organization and keyword search. In the middle—where most innovation clusters—LLM-enhanced tools like Obsidian with Smart Connections, Notion AI, and Reflect bolt semantic search, summarization, and chat interfaces onto existing structures. The workflows remain fundamentally unchanged; AI just accelerates them.

At the AI-native extreme sits Mem.ai, which represents a genuine philosophical departure. Rather than asking users to organize, it assumes AI should handle that entirely. There are no folders. Connections emerge from AI analysis. Retrieval happens through natural language conversation. This approach trades user control for automated intelligence—a tradeoff that divides the PKM community.

Notion’s September 2025 release of AI Agents marks another significant shift: autonomous multi-step task execution that can update hundreds of database pages, build project plans, and work across connected tools for up to twenty minutes without human intervention. This moves from “AI that suggests” to “AI that executes”—blurring the line between assistant and worker.

The common implementation pattern across tools is RAG (Retrieval-Augmented Generation): notes get chunked into segments, converted to vector embeddings, indexed in a database, and retrieved via semantic similarity when queried. This technical architecture enables natural language querying of personal knowledge bases but introduces new limitations—context window constraints mean “chat with everything” often means “chat with a retrieved subset.”

Emerging AI-Native Tools:

Reor - A private, local AI PKM app designed for “high entropy people,” running entirely offline with local models
Tana - Combines supertags with AI integration, offering voice chat for idea development plus semantic connection-finding
InfraNodus - Imports vaults, visualizes as network graphs, uses AI to identify structural gaps and generate bridging ideas

Thought Leader Adaptations: Additive Rather Than Transformative

Tiago Forte (Building a Second Brain / CODE)

Forte argues PKM becomes more valuable with AI, not obsolete. His May 2024 position: “No matter how powerful AI becomes, the data we put into it has to come from somewhere, and the AI’s outputs have to go somewhere.”

He’s reframed CODE around AI’s impact on each stage:

Capture remains human-driven
Organize is “radically transformed” by AI
Distill is a “perfect fit for AI” (summarization)
Express “still requires a human” for voice and perspective

His key insight: AI concentrates human creativity at the beginning and end of the creative process, handling the middle.

Sönke Ahrens (How to Take Smart Notes / Zettelkasten)

Ahrens takes a more philosophical stance. His view: if you understand thinking as happening externally as well as internally, AI becomes a tool rather than a replacement. The Zettelkasten serves as a “dialogue partner,” and these personal external thinking spaces become more critical in the AI era.

His warning: “If you think we only think with our brains, you tend to see AI as a replacement of our thinking process… and that’s obviously a risk.”

Nick Milo (Linking Your Thinking)

Milo has fully embraced AI, adding an “AI + PKM Masterclass” to his LYT Workshop. His framing: treat AI chatbots as “interns” (borrowing from Ethan Mollick) for brainstorming enhancement rather than note replacement.

David Allen (GTD)

Allen’s core framework remains unchanged. GTD’s five steps (Capture, Clarify, Organize, Reflect, Engage) continue as described, with AI tools serving as better capture mechanisms and smarter organizational assistants rather than workflow restructuring.

The notable pattern: every major thought leader has added AI content to their offerings, but none has fundamentally restructured their methodology. Adaptations are additive rather than transformative. GTD remains a five-step workflow; PARA still has four folders; Zettelkasten still demands atomic notes and manual linking. The gap between “AI-enhanced traditional PKM” and “AI-native PKM” remains largely unexplored by established voices.

Cognitive Science: The “AI Assistance Dilemma”

Academic research presents a sobering counterweight to AI enthusiasm.

Critical Thinking Correlation

A 2025 study of 666 participants found a strong negative correlation (r = -0.68) between AI tool usage and critical thinking abilities, with younger users (17-25) exhibiting higher AI dependence and lower critical thinking scores. Higher education served as a protective buffer.

The Preference-Performance Gap

Microsoft Research compared LLM use versus note-taking in 405 students: note-taking alone or combined with LLMs had significant positive effects on retention compared to LLM alone. Yet students preferred using LLMs and perceived them as more helpful—despite worse outcomes. This preference-performance gap represents a core design challenge.

The Inverted U-Curve of AI Assistance

A 2025 CSCW study on AI-assisted note-taking found an inverted U-curve:

Automated AI (high assistance) produced the lowest learning outcomes despite highest perceived ease of use
Intermediate AI (moderate assistance) achieved the highest learning outcomes
Manual (no AI) fell in the middle

This aligns with Robert and Elizabeth Bjork’s research on “desirable difficulties”—learning challenges that feel harder but produce better retention.

Human-AI Collaboration Research

MIT’s Center for Collective Intelligence found that on average, AI-human combinations do not outperform the best human-only or AI-only systems—except for creation tasks, where positive synergy emerged.

Key mechanisms for effective collaboration:

AI delegation - AI passing tasks to humans in relevant contexts
Capability complementarity - Leveraging non-overlapping strengths
Explicit human oversight of synthesis and judgment

The implication for PKM design: the most effective AI integration will preserve beneficial friction while automating genuine toil. What feels easiest isn’t what builds understanding. Systems that scaffold thinking without substituting it outperform both pure manual effort and full automation.

Community Discourse: Obsolescence vs. Essential

The “Zettelkasten is Obsolete” Movement

A growing faction argues traditional PKM has become productivity theater. Ev Chapman’s influential critique: “Luhmann built his system in an analog world. No search functions, no AI, no ability to instantly find connections across thousands of notes. All that manual organization & linking? It was necessary 50 years ago. Today, it’s just busywork.”

The Counter-Argument

Practitioners like Sascha (Zettelkasten.de founder) counter: “The effort of doing these things is tiring and takes time, but it’s thinking itself. Using AI for thinking activities is like using a motorcycle during training for a marathon.”

The core question: does semantic search eliminate the need for atomic notes and careful linking, or does the process of creating links with full sentences explaining connections constitute the thinking itself?

Emerging Novel Frameworks

Self-Organizing Second Brain (Chris Lettieri): Ideas cluster organically based on semantic similarity; AI auto-distills related concepts into unified notes
Constella’s “Real-Time AI Recall”: As you type, the system surfaces related past notes proactively
Tana’s Supertag + AI Integration: Voice chat for idea development plus semantic connection-finding across notes
InfraNodus PKM Workflow: Import vaults, visualize as network graph, use AI to identify structural gaps and generate bridging ideas

The Biggest Discourse Gap

No one is asking what’s lost when AI assists synthesis. The community celebrates AI help but rarely examines cognitive costs. Zettelkasten’s value comes partly from the struggle of making connections. AI connections may be more comprehensive but less surprising. Is human-discovered insight more valuable than AI-surfaced patterns?

Positioning “Ground → Capture → Think → Recall”

Mapping Against Existing Frameworks

Phase	Existing Analog	What’s Different
Ground	No direct equivalent	Explicit pre-research before capture is novel
Capture	GTD inbox, CODE Capture, Zettelkasten fleeting notes	Standard—well-established pattern
Think	CODE Distill, Zettelkasten processing	AI conversation as synthesis medium is new
Recall	Traditional retrieval + emerging semantic search	Hybrid human/AI retrieval is AI-native

What Makes It Distinct

The “Ground” step is the most distinctive contribution. Existing frameworks assume you encounter information then capture it. GTD starts with “stuff” already in your life. CODE’s first step is Capture. Zettelkasten begins with reading, but reading isn’t systematized as a preceding “grounding” activity.

The explicit separation of “understand context” from “capture ideas” creates space for AI-assisted research before personal synthesis—recognizing that modern knowledge work often requires understanding a landscape before contributing to it.

“Think via AI Conversations” represents a fundamentally different synthesis method. Traditional PKM synthesis is solo, iterative, and asynchronous—you read, extract, link, and process alone over time. AI-conversation synthesis is dialogic, real-time, and collaborative. The thinking happens during conversation, not after. This enables processing more volume while maintaining quality but creates risk of reduced deep personal engagement with ideas.

Spectrum Position

The pattern positions itself firmly in the AI-native tier—not merely adding AI features to existing structures but designing around conversational interfaces and emergent organization from the start.

Critical Gaps: Questions the Community Hasn’t Addressed

How should capture change when AI can retrieve anything? Traditional capture assumes forgetting; semantic search changes this calculus. Should we capture less, capture differently, or capture with different metadata? The “capture everything” versus “be selective” debate needs AI-era revisiting.
What happens to emergence when connections are AI-suggested? Zettelkasten’s value comes from unexpected connections produced by human linking. AI connections may be more comprehensive but less surprising. The trade-off between serendipity and completeness is underexamined.
How do we maintain epistemic agency with AI partners? When AI synthesizes, whose knowledge is it? What’s the provenance trail for co-created insights? How do we verify AI-assisted understanding against ground truth?
What’s the right “Ground” for different knowledge work? The Ground step implies context matters—but how much? When? Should grounding be standardized or situation-dependent? This is largely unexplored.
How does AI change what the “unit of knowledge” should be? Notes have been the assumed unit; conversations may be equally valuable. AI enables “synthesis on demand” rather than “synthesis then retrieval”—potentially inverting the traditional workflow entirely.

Conclusion: Design for Cognitive Partnership, Not Cognitive Replacement

The future of PKM isn’t choosing between human effort and AI automation—it’s designing workflows where each handles what it does best:

AI excels at:

Information retrieval
Pattern recognition across large corpora
Semantic understanding
Surfacing non-obvious connections

Humans excel at:

Judgment
Meaning-making
Epistemic evaluation
Creative synthesis requiring context, stakes, and values

The “Ground → Capture → Think → Recall” pattern represents a genuinely AI-native approach by front-loading research (Ground), maintaining low-friction intake (Capture), centering dialogic synthesis (Think), and enabling hybrid retrieval (Recall). Its novelty lies primarily in the explicit grounding phase and the treatment of AI conversation as a primary thinking medium rather than a retrieval tool.

Key Design Principles for AI-Era PKM

Preserve “desirable difficulties” that build understanding
Treat AI as thinking partner in synthesis, not replacement for thinking
Separate information gathering (AI-suitable) from meaning-making (human-essential)
Design for intermediate assistance—not minimal and not maximal
Maintain clear epistemic ownership and provenance
Front-load context before capture when the knowledge domain is unfamiliar

The practitioners who will thrive aren’t those who ignore AI or those who delegate thinking entirely—they’re those who build workflows that leverage AI’s capabilities while preserving the cognitive engagement that makes knowledge work valuable in the first place.

Sources

ResearchGate - “AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking” https://www.researchgate.net/publication/387701784_AI_Tools_in_Society_Impacts_on_Cognitive_Offloading_and_the_Future_of_Critical_Thinking
Aiappgenie - “Mem AI Tool Review 2025: Features, Use Cases, Pricing, and Top Alternatives” https://aiappgenie.com/post/mem-ai-tool
Skywork - “The 8 Best AI Note-Taking Apps to Build Your Second Brain (2025)” https://skywork.ai/blog/the-8-best-ai-note-taking-apps-to-build-your-second-brain-2025/
Notion - “September 18, 2025 – Notion 3.0: Agents” https://www.notion.com/releases/2025-09-18
Max Productive AI - “Notion AI Review 2025: Features, Pricing & AI Agents Guide” https://max-productive.ai/ai-tools/notion-ai/
GitHub - “reorproject/reor: Private & local AI personal knowledge management app for high entropy people” https://github.com/reorproject/reor
Forte Labs - “Will Artificial Intelligence Replace the Need for Second Brains Entirely?” https://fortelabs.com/blog/will-artificial-intelligence-replace-the-need-for-second-brains-entirely/
Mindhack Podcast - “AI-Proof Your Brain: Master Smart Notes with Sönke Ahrens” https://mindhack.com/episode/094-snke-ahrens-ai-proof-your-brain-smart-notes-strategy-thinkwithnotes
Scott Loftesness - “Reflections on the Linking Your Thinking Workshop” https://sjl.us/2023/07/16/reflections-on-the-linking-your-thinking-workshop/
Workflowy - “Productivity Masters: David Allen – A Q&A With the Creator of the GTD system” https://blog.workflowy.com/productivity-masters-david-allen-gtd/
IEEE Computer Society - “The Personalized Learning Revolution” (Cognitive Offloading) https://www.computer.org/publications/tech-news/trends/cognitive-offloading
Microsoft Research - “Effects of LLM Use and Note-Taking On Reading Comprehension and Memory: A Randomised Experiment in Secondary Schools” https://www.microsoft.com/en-us/research/publication/effects-of-llm-use-and-note-taking-on-reading-comprehension-and-memory-a-randomised-experiment-in-secondary-schools/
arXiv - “More AI Assistance Reduces Cognitive Engagement: Examining the AI Assistance Dilemma in AI-Supported Note-Taking” https://arxiv.org/html/2509.03392v1
Funblocks - “Desirable Difficulty” https://www.funblocks.net/thinking-matters/classic-mental-models/desirable-difficulty
Education Learning Pro - “Why Struggle Makes You Smarter: Understanding Desirable Difficulties in Learning” https://educationlearningpro.com/why-struggle-makes-you-smarter-understanding-desirable-difficulties-in-learning/
MIT Sloan - “When humans and AI work best together — and when each is better alone” https://mitsloan.mit.edu/ideas-made-to-matter/when-humans-and-ai-work-best-together-and-when-each-better-alone
The Decision Lab - “Human-AI Collaboration” https://thedecisionlab.com/reference-guide/computer-science/human-ai-collaboration
Ev Chapman - “Why I Ditched Zettelkasten for a Centuries Old Note-Taking System (& you should too)” https://www.evchapman.com/blog/why-i-ditched-zettelkasten-for-a-centuries-old-note-taking-system-you-should-too
Zettelkasten Forum - “Are AI and Zettelkasten compatible each other?” https://forum.zettelkasten.de/discussion/2863/are-ai-and-zettelkasten-compatible-each-other
Medium (Chris Lettieri) - “Self-Organizing Second Brain: How I Manage Information Overload” https://medium.com/@BitsOfChris/self-organizing-second-brain-how-i-manage-information-overload-2266bd0d9e27
Medium (Theo James) - “Use These ChatGPT Prompts to build your PKMS” https://medium.com/@theo-james/use-these-chatgpt-prompts-to-build-your-pkms-6a406e6af9bf
Medium (Ann P.) - “Your Second Brain Is Broken: Why Most PKM Tools Waste Your Time” https://medium.com/@ann_p/your-second-brain-is-broken-why-most-pkm-tools-waste-your-time-76e41dfc6747
Nodus Labs - “PKM Workflow: AI-generated Insights for Your Obsidian / LogSeq Knowledge Graphs” https://support.noduslabs.com/hc/en-us/articles/6455436092690—PKM-Workflow-AI-generated-Insights-for-Your-Obsidian-LogSeq-Knowledge-Graphs
Hamberg - “GTD in 15 minutes – A Pragmatic Guide to Getting Things Done” https://hamberg.no/gtd
Workflowy - “Build a second brain - Workflowy guide” https://workflowy.com/systems/build-a-second-brain/

PKM + LLM Integration Research