llm_wiki_second_brain_augmented_human_research_2026-05-24

LLM Wiki / Second Brain As Augmented Human

Date: 2026-05-24 Status: research note for Memova positioning

Short Answer

There is no serious academic consensus that an LLM-wiki user becomes a different biological species or literally gains infinite memory / superhuman IQ.

There is, however, a strong academic and AI-practitioner tradition that supports a slightly different claim:

A person with a reliable, always-available, agent-readable second brain becomes a different cognitive system from the same person without it.

The defensible framing is not "new organism" but extended cognitive system, human-computer symbiosis, or augmented cognition.

Strongest Support

Famous Cases

Niklas Luhmann's Zettelkasten

Best Memova-relevant case.

German sociologist Niklas Luhmann built a Zettelkasten of more than 90,000 linked index cards. The German Akademienunion project describes these cards as the central object of the research project "Niklas Luhmann - Theorie als Passion." Luhmann is the cleanest historical example of a person whose productivity and theory-building were inseparable from a structured external memory system.

Why it matters:

• It was not a passive archive.
• It was a thinking partner / communication partner.
• It supported long-horizon synthesis across sociology, law, systems theory, media, art, religion, ecology, and organization.

Useful sources:

• Akademienunion, "90.000 Zettel": https://www.akademienunion.de/akademienforschung/90000-zettel
• Niklas Luhmann Archive: https://niklas-luhmann-archiv.de/

Charles Darwin's Notebooks

Darwin's "Transmutation Notebooks" are famous because they capture the early formation of evolutionary theory. Notebook B includes the iconic "Tree of Life" sketch and the phrase "I think." Cambridge materials and later coverage identify these notebooks as where Darwin first theorized species transmutation.

Why it matters:

• The notebook was the place where evidence, speculation, classification, and theory formation interacted.
• The theory was not born only inside Darwin's head; it emerged through a repeated note-making process over years.

Useful sources:

• Cambridge Digital Library Darwin manuscripts: https://cudl.lib.cam.ac.uk/collections/darwin_mss
• Euronews / Cambridge coverage of returned Darwin notebooks: https://www.euronews.com/2022/04/05/uk-britain-notebooks-darwin
• Darwin's Other Books, PNAS / PMC: https://pmc.ncbi.nlm.nih.gov/articles/PMC1283389/

Thomas Edison's Laboratory Notebooks

Edison's lab is one of the best-known industrial R&D examples of externalized memory. His notebooks recorded experiments, sketches, failures, patent-relevant dates, and collaborative lab work. They were not just personal notes; they were an operating memory for an invention organization.

Why it matters:

• The notebook becomes an institutional memory layer.
• It supports reproducibility, priority claims, iteration, and collaboration.
• This is close to Memova's "project memory + action history" framing.

Useful sources:

• Edison Papers, Rutgers: https://edisondigital.rutgers.edu/
• National Park Service, Thomas Edison National Historical Park collections: https://www.nps.gov/edis/learn/historyculture/collections.htm

Richard Feynman's Notebooks

Feynman is useful because he resisted the idea that notebooks merely record thinking after the fact. In an interview about his notebooks, he objected to the phrase "record of your day-to-day work," saying he actually did the work on the paper.

Why it matters:

• The notebook is not memory after thinking.
• The notebook is part of thinking.
• This is a strong line for Memova: "the page is not where thoughts are stored; it is where thoughts become operable."

Useful source:

• Notebook interview excerpt used in science notebook pedagogy materials: https://www3.nd.edu/~nismec/hasti2013/hasti2013-bellina-notebooks.pdf

Vannevar Bush's Memex

Vannevar Bush's 1945 "As We May Think" proposed the memex: a personal machine for storing books, records, communications, annotations, and associative trails. This is one of the foundational computer-science visions of external memory and hypertext.

Why it matters:

• It explicitly frames technology as a memory extender.
• It anticipates personal knowledge bases, hypertext, backlinks, and linked trails.
• It is the historical ancestor of an LLM wiki.

Useful source:

• Vannevar Bush, "As We May Think": https://www.theatlantic.com/magazine/archive/1945/07/as-we-may-think/303881/

Douglas Engelbart's Augmentation System

Engelbart's work on augmenting human intellect treated computers, displays, documents, links, and collaborative work systems as tools for increasing collective intellectual capability. His NLS demo is a canonical example of computers as thinking infrastructure.

Why it matters:

• It turns external memory into action and collaboration.
• It is less "note app" and more "operating system for thought."
• This is close to Memova's notes -> memory -> actions arc.

Useful source:

• Douglas Engelbart, "Augmenting Human Intellect": https://www.dougengelbart.org/content/view/138/

Licklider's Human-Computer Symbiosis

Licklider's 1960 paper is not a note-taking case, but it is a foundational AI/computer-science case for human-computer coupling. It imagines humans and computers thinking together, each doing what the other does poorly.

Why it matters:

• Second brain + agent is exactly a symbiosis architecture.
• Human judgment and goals remain central; machine memory and transformation extend the person.

Useful source:

• J. C. R. Licklider, "Man-Computer Symbiosis": https://www.columbia.edu/~jrh29/licklider/man-computer_symbiosis.html

Karpathy's LLM Wiki

Karpathy's LLM Wiki pattern is the modern AI-world articulation of a persistent, compounding wiki maintained by LLMs. It contrasts with stateless chat and shallow RAG.

Why it matters:

• The wiki is a durable substrate for agents.
• It compounds across sessions.
• It turns the user's memory into an operational context layer.

Useful source:

• Andrej Karpathy, "LLM Wiki" gist: https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f

Michael Nielsen / Andy Matuschak

Michael Nielsen's "Augmenting Long-term Memory" and Andy Matuschak's evergreen notes represent the modern AI-adjacent research/productivity tradition: memory systems should be designed to change what a person can understand, recall, and build over time.

Why it matters:

• They connect notes, spaced repetition, research, and cognition.
• They show that external memory is not just retrieval; it can reshape learning and theory-building.

Useful sources:

• Michael Nielsen, "Augmenting Long-term Memory": https://augmentingcognition.com/
• Andy Matuschak notes: https://notes.andymatuschak.org/

Modern AI Enterprise Examples

Modern AI Founder / Operator Cases

The number of famous AI founders who publicly reveal a detailed personal second-brain workflow is smaller than the hype suggests. Most executives talk about enterprise memory or AI agents, but do not publish their own personal knowledge system. The strongest public personal/operator cases are below.

Andrej Karpathy: LLM Wiki

Strongest modern AI figure case.

Karpathy's 2026 LLM Wiki gist describes a personal knowledge base maintained by an LLM: raw sources are immutable, the LLM maintains a structured markdown wiki, and a schema file tells the agent how to ingest, update, query, and lint the wiki. He explicitly says he runs the LLM agent on one side and Obsidian on the other; "Obsidian is the IDE; the LLM is the programmer; the wiki is the codebase."

Why it matters:

• Karpathy is a highly visible AI builder: OpenAI cofounder, former Tesla AI lead, Eureka Labs founder.
• The system is not a note-taking productivity trick; it is an agent-readable memory architecture.
• It became a template for many LLM-wiki / Obsidian / Claude Code implementations.

Source:

• Andrej Karpathy, LLM Wiki gist: https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f

Farza Majeed: Farzapedia

Farza, founder of buildspace, built "Farzapedia" by feeding about 2,500 diary entries, Apple Notes, and iMessage conversations into an LLM, which compiled them into roughly 400 interlinked personal wiki articles. Coverage describes the result as being built for the agent, not primarily for Farza to browse manually.

Why it matters:

• This is one of the most concrete founder/operator examples of turning personal memory into agent-readable infrastructure.
• It uses life/work context, not only research PDFs.
• It shows why an explicit wiki can be more useful than raw chat logs or generic RAG: the agent can crawl the map.

Sources:

• Toolmesh, Karpathy's LLM Wiki / Farzapedia: https://www.toolmesh.ai/news/karpathy-llm-wiki-personal-knowledge-management
• Floatboat, What is LLM Wiki / Farzapedia: https://floatboat.ai/blog/what-is-llm-wiki
• Crunchbase, Farza Majeed / buildspace: https://www.crunchbase.com/person/farza-majeed-2063

Jeremy Howard / fast.ai / Answer.AI: Notebooks As Research-Software Memory

Jeremy Howard, fast.ai cofounder and Answer.AI founder, is not usually framed as "second brain," but his tooling culture is notebook-first. fast.ai's nbdev makes Jupyter notebooks the place where software, tests, documentation, and exploration live together.

Why it matters:

• For AI researchers/builders, notebooks are a classic external memory: experiment, code, narrative, result, and explanation in one artifact.
• nbdev turns notebooks from scratchpads into maintainable software memory.
• This is the engineering equivalent of "the notebook is where thinking becomes executable."

Sources:

• nbdev tutorial: https://fastai.github.io/jupytercon-2023/
• fastai paper: https://arxiv.org/abs/2002.04688

Simon Willison: Public TIL / Blog As External Memory For AI Engineering

Simon Willison, Datasette founder and a major LLM tooling voice, maintains a public blog and TIL system. This is not an LLM-maintained second brain in the Karpathy sense, but it is a famous developer example of externalized technical memory: small, searchable notes that accumulate into a public working memory.

Why it matters:

• His TIL practice influenced many developers to externalize small learnings instead of relying on recall.
• His LLM writing often shows a workflow of turning experiments into durable notes, then reusing them as public knowledge.
• This is weaker than Karpathy/Farzapedia as an agent-memory case, but useful as AI-engineering culture.

Source:

• Simon Willison, "One year of TILs": https://simonwillison.net/2021/May/2/one-year-of-tils/

Dhravya Shah / Supermemory

Dhravya Shah founded Supermemory, described by TechCrunch as a universal memory API for AI apps. This is more of a startup case than a personal workflow case, but it shows young AI founders treating memory as an independent infrastructure layer rather than a feature.

Why it matters:

• It validates the "memory layer for AI apps" category.
• It is closer to infrastructure than note-taking.

Source:

• TechCrunch, Supermemory: https://techcrunch.com/2025/10/06/a-19-year-old-nabs-backing-from-google-execs-for-his-ai-memory-startup-supermemory/

Caution

Do not overclaim that Sam Altman, Dario Amodei, Demis Hassabis, Elon Musk, or other frontier-lab CEOs have famous public second-brain workflows. They often talk about agents, intelligence, memory, or enterprise context, but there is not enough public evidence that they personally use an explicit LLM wiki / second brain in the Karpathy/Farza sense.

OpenAI: ChatGPT Projects / Project-Only Memory / Codex

OpenAI's Projects make context persistent around a body of work: chats, files, instructions, and related context stay together so users do not need to restate background every time. OpenAI also documents project-only memory, where context stays inside a project boundary.

Why it matters:

• OpenAI is moving from stateless chat toward durable workspaces.
• "Project" becomes the memory boundary.
• Codex extends this pattern into repositories, tasks, and longer-running agent work.

Useful sources:

• OpenAI Academy, "Using projects in ChatGPT": https://openai.com/academy/projects/
• OpenAI, "Scaling Codex to enterprises worldwide": https://openai.com/index/scaling-codex-to-enterprises-worldwide/

Google: Gemini Notebooks / NotebookLM Enterprise

Google introduced Notebooks in Gemini as personal knowledge bases synced with NotebookLM. Google explicitly frames Notebooks as a way to organize chats and files for complex projects, give Gemini context, and avoid re-explaining the project.

Why it matters:

• Google is turning notebook/project context into the shared memory layer for Gemini and NotebookLM.
• NotebookLM Enterprise brings this pattern into enterprise search and document workflows.

Useful sources:

• Google, "Try notebooks in Gemini": https://blog.google/innovation-and-ai/products/gemini-app/notebooks-gemini-notebooklm/
• Google Cloud, NotebookLM Enterprise within Gemini Enterprise: https://docs.cloud.google.com/gemini/enterprise/docs/access-notebooklm

Microsoft: SharePoint Agents / Microsoft Graph / Permissioned Enterprise Memory

Microsoft's SharePoint agents are grounded in sites, pages, document libraries, folders, files, and Microsoft Graph. Microsoft emphasizes that agents only expose content the user already has permission to access.

Why it matters:

• This is enterprise second brain as permissioned organizational memory.
• The key moat is not only model intelligence; it is identity, access, document graph, and governance.

Useful source:

• Microsoft Support, FAQ about Copilot in SharePoint: https://support.microsoft.com/en-gb/office/frequently-asked-questions-about-copilot-in-sharepoint-eb1b7668-3d98-4a93-98ef-f0c6dfc694f0

Anthropic: Claude Enterprise / Connectors / Plugins / Cowork

Anthropic's enterprise push positions Claude as a subject-matter expert over internal knowledge and as an agent that connects to systems across finance, engineering, design, legal, HR, and enterprise tools. Its enterprise direction centers on connectors, plugins, governance, and workflow context.

Why it matters:

• Anthropic's enterprise agent thesis is that model capability must be paired with company-specific context.
• Connectors/MCPs turn scattered enterprise systems into usable agent memory.

Useful sources:

• Anthropic Enterprise: https://www.anthropic.com/enterprise
• TechCrunch, Anthropic enterprise agents and plug-ins: https://techcrunch.com/2026/02/24/anthropic-launches-new-push-for-enterprise-agents-with-plugins-for-finance-engineering-and-design/

Glean: Enterprise Graph

Glean is one of the clearest enterprise examples. Its Enterprise Graph captures relationships across company data, people, projects, teams, and processes, combining organization-wide knowledge with personal graphs.

Why it matters:

• This is basically a company-level second brain for agents.
• Glean's core insight is that enterprise AI depends on knowledge graph + permissions + connectors, not just a chat UI.

Useful source:

• Glean Enterprise Graph: https://www.glean.com/product/enterprise-graph

Hebbia: Matrix For Finance / Legal Knowledge Work

Hebbia's Matrix is a multi-agent platform for complex finance and legal workflows over documents, spreadsheets, contracts, filings, and presentations. OpenAI frames Hebbia as automating high-value knowledge work with agents that analyze documents and generate validated artifacts.

Why it matters:

• It shows enterprise memory in action: large document collections become the substrate for agent workflows.
• The workflow is not "ask a chatbot"; it is document corpus -> analysis -> validation -> artifact.

Useful sources:

• OpenAI, "Automating 90% of finance and legal work with agents": https://openai.com/index/hebbia/
• Hebbia Product: https://www.hebbia.com/product

Notion: Workspace As Agent Hub

Notion's Custom Agents turn the workspace itself into an agent substrate. Notion's direction is especially relevant because it already owns notes, docs, databases, projects, meeting notes, and company knowledge in one place.

Why it matters:

• Notion is a consumer/enterprise bridge for "workspace as second brain."
• Custom agents make structured workspace data executable.

Useful sources:

• Notion 3.3 Custom Agents release: https://www.notion.com/releases/2026-02-24
• TechCrunch, Notion as hub for AI agents: https://techcrunch.com/2026/05/13/notion-just-turned-its-workspace-into-a-hub-for-ai-agents/

mem0 / Letta: Memory Infrastructure For Agents

mem0 and Letta are not consumer note apps; they are developer infrastructure for agent memory and stateful agents. mem0 frames itself around scalable long-term memory for production AI agents. Letta similarly focuses on stateful agents with memory.

Why it matters:

• The developer ecosystem has converged on memory as a first-class agent primitive.
• This supports the Memova thesis that memory is infrastructure, not a feature.

Useful sources:

• mem0 paper: https://arxiv.org/abs/2504.19413
• mem0 GitHub: https://github.com/mem0ai/mem0
• Letta GitHub: https://github.com/letta-ai/letta

Extended Mind

Clark and Chalmers' 1998 "The Extended Mind" argues that cognition can extend beyond the skull when an external artifact plays the same functional role as internal cognition. The canonical case is Otto's notebook: if Otto reliably uses a notebook the way another person uses biological memory, the notebook can be treated as part of Otto's memory system.

Relevance to LLM wiki:

• A well-maintained LLM wiki is not just storage.
• It becomes a structured, accessible, action-guiding memory layer.
• If it is reliable, constantly available, and integrated into behavior, it can be described as part of the user's extended cognitive system.

Useful source:

• Andy Clark and David Chalmers, "The Extended Mind": https://www.consc.net/papers/extended.html

Human-Computer Symbiosis

J. C. R. Licklider's 1960 "Man-Computer Symbiosis" predicted tightly coupled human-computer systems that would think and process data more effectively than either ordinary humans or conventional machines alone.

Relevance to Memova:

• Humans set goals, hypotheses, values, and evaluations.
• Computers handle memory, retrieval, transformations, and routine operations.
• This is very close to agent + second brain: the user remains the intentional center, while the machine expands memory and execution.

Useful source:

• J. C. R. Licklider, "Man-Computer Symbiosis": https://www.columbia.edu/~jrh29/licklider/man-computer_symbiosis.html

Engelbart / Intelligence Amplification

Douglas Engelbart framed computers as a way to augment human intellect, especially through external symbol systems. His examples include external memory and visualization structures for concepts the mind cannot reliably remember or manipulate alone.

Relevance to LLM wiki:

• The wiki is an external symbol structure.
• The LLM maintains links, contradictions, summaries, and retrieval paths.
• The system increases intellectual effectiveness, especially on long-running, complex projects.

Useful source:

• Douglas Engelbart, "Augmenting Human Intellect": https://www.dougengelbart.org/content/view/138/

LLM Wiki / Knowledge Compounding

Karpathy's LLM Wiki pattern explicitly describes a persistent, compounding wiki maintained by an LLM rather than a stateless RAG system. The wiki accumulates synthesis, cross-references, contradictions, and decisions over time.

Relevance to Memova:

• A normal chatbot session is temporary.
• A wiki maintained by an agent compounds.
• The user becomes a person plus a persistent, source-grounded memory/codebase.

Useful source:

• Andrej Karpathy, "LLM Wiki" gist: https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f

Functional Infinite Context

Recent AI-memory work frames active external memory as a way to move beyond passive retrieval toward "functional infinite context" and cognitive augmentation. This does not mean literally infinite memory, but it supports the idea that structured external memory changes the capability profile of the human-agent system.

Useful sources:

• Cognitive Workspace: Active Memory Management for LLMs: https://arxiv.org/abs/2508.13171
• WiCER: Wiki-memory Compile, Evaluate, Refine: https://arxiv.org/abs/2605.07068

Balanced View

The strongest claim Memova can make:

People who build and use an agent-readable second brain are not merely using better notes. They are operating as extended cognitive systems: their memory, retrieval, synthesis, and execution loops partly live outside the biological brain.

The claim to avoid:

These people have infinite memory or superhuman IQ.

Better wording:

• "functionally much larger memory"
• "persistent, searchable, source-grounded memory"
• "higher effective intelligence in long-horizon workflows"
• "agent-augmented cognition"
• "extended mind for everyday work"

Risks / Counterpoints

Second-brain systems can also weaken cognition if the user offloads judgment, not just memory. Anthropic's disempowerment-patterns research and broader cognitive-offloading literature warn about deferential AI use, reality distortion, automation bias, and loss of critical engagement.

Memova implication:

• The product should not say "let AI think for you."
• It should say "capture your context, preserve your judgment, and let agents act with your memory."
• Human remains the decision-maker; second brain is the alignment and recall layer.

Useful sources:

• Anthropic, "Disempowerment patterns in real-world AI usage": https://www.anthropic.com/research/disempowerment-patterns
• Stack Overflow, "AI is becoming a second brain at the expense of your first one": https://stackoverflow.blog/2026/03/19/ai-is-becoming-a-second-brain-at-the-expense-of-your-first-one/

Memova Positioning Line

Chinese:

有第二大脑的人不是生物学意义上的新人类，但在认知系统意义上已经和普通 AI 用户不同：他的记忆、上下文、关系、项目状态和行动历史都被结构化成 agent 可读取、可验证、可持续更新的外部大脑。

English:

A person with an agent-readable second brain is not a new biological species, but they are a different cognitive system: memory, context, relationships, project state, and action history become external, durable, searchable, and executable by agents.

Workflow Shift: Pre-AI / AI Era vs Agent + LLM Wiki Era

Old Workflow: Human Brain As The Bus

The pre-agent workflow turns the human into the integration layer.

1. Inbox

• Inputs arrive from email, chat, meetings, documents, browser tabs, photos, books, podcasts, handwritten notes, and tasks.
• Everything is fragmented across tools.

1. Capture

• User saves links, takes notes, screenshots, highlights, bookmarks, or meeting notes.
• Capture is lossy and inconsistent.

1. Triage

• User manually decides what matters, what to ignore, what to tag, what to move, and what to remember.
• Most information dies here.

1. Storage

• Notes go into folders, docs, Notion, Obsidian, Apple Notes, Google Drive, Slack, email, or local files.
• Storage helps future search, but it does not understand the user's projects.

1. Recall

• User must remember that something exists, where it lives, what words to search, and why it mattered.
• Search returns documents, not decisions.

1. Synthesis

• User manually reads, compares, extracts, and connects ideas.
• This is where expert productivity lives, but it is also the bottleneck.

1. Planning

• User turns synthesis into todos, outlines, specs, slides, emails, code tasks, or decisions.
• The plan is usually separated from the source context.

1. Production

• User creates the final output: memo, pitch, code, email, design, paper, document, decision, or meeting action.
• Output rarely updates the memory system cleanly.

1. Feedback / Archive

• User may file the result somewhere, but the next cycle often starts from scratch.

Core problem:

The old workflow stores information, but the human remains the router, memory, synthesizer, planner, and executor.

ChatGPT Era Workflow: Better Assistant, Still Stateless

The chatbot era improves synthesis and drafting but still depends on the user to carry context.

1. Inbox stays fragmented.
2. User manually selects context.
3. User pastes snippets into chat.
4. LLM summarizes, rewrites, or drafts.
5. User manually checks the answer.
6. User manually moves the output back into docs, tasks, email, code, or slides.
7. The next session forgets most of the structure unless the user re-explains it.

Core problem:

ChatGPT helps with local tasks, but the user is still the memory bridge between conversations, tools, projects, and outputs.

Agent + LLM Wiki Workflow: Memory As The Bus

The agent era changes the architecture. The wiki/memory layer becomes the integration surface.

1. Inbox

• Inputs still arrive from many places: notes, meetings, email, chat, docs, browser, code, calendar, tasks, voice, handwriting, and photos.
• But the system treats them as memory events, not random files.

1. Capture

• Agent captures raw sources with provenance: timestamp, source, people, project, permissions, original file/link/transcript.
• Raw data stays separate from synthesized memory.

1. Ingestion

• Agent extracts entities, commitments, questions, decisions, contradictions, relationships, and action candidates.
• The user does not need to manually tag everything.

1. Wiki / Memory Update

• Agent updates canonical pages: people, projects, decisions, meetings, tasks, open questions, product facts, user preferences.
• The memory base compounds instead of becoming a pile.

1. Retrieval

• Agent retrieves context by project, person, intent, timeline, source, and decision history.
• Search becomes contextual recall, not keyword hunting.

1. Synthesis

• Agent compares sources, summarizes changes, resolves conflicts, points out uncertainty, and proposes next steps.
• Human reviews and corrects judgment, not every raw connection.

1. Planning

• Agent turns memory into plans: todo lists, meeting briefs, product specs, slide outlines, code-change plans, research maps, follow-ups.
• Plans are linked back to evidence.

1. Tool Execution

• Agent can operate tools: create docs, draft emails, update tasks, edit code, generate slides, schedule calendar events, query databases.
• Human confirms high-impact actions.

1. Output

• Final output is produced with source-grounded context: memo, pitch deck, code, report, email, meeting action, investor narrative, product decision.

1. Feedback / Memory Writeback

• The final output, user corrections, decisions, and next actions are written back into the wiki.
• The next cycle starts with better memory than the previous cycle.

Core shift:

In the agent + LLM wiki era, the memory layer becomes executable. The output is no longer the end of the workflow; it becomes new context for future work.

Comparison Table

Representative Products By Workflow Stage

The older workflow already had strong products at every stage. The gap was not lack of tools; the gap was that the human had to connect them.

Product-Category Interpretation

Pre-AI products were excellent at single-stage jobs:

• Gmail / Slack owned inbox.
• Evernote / OneNote / Apple Notes owned capture.
• Dropbox / Drive owned storage.
• Notion / Confluence owned knowledge pages.
• Todoist / Asana / Jira owned tasks.
• Docs / PowerPoint / Figma / GitHub owned outputs.

But the user remained the workflow glue.

ChatGPT-era products improved local synthesis and drafting:

• ChatGPT / Claude / Gemini helped summarize and generate.
• Perplexity improved research and answer retrieval.
• NotebookLM helped reason over uploaded sources.
• Cursor / Copilot helped inside code.
• Notion AI / Google Workspace AI / Microsoft Copilot helped inside documents.

But most outputs still had to be manually moved across tools, and long-term memory remained fragmented.

Agent + LLM wiki products are beginning to own the cross-stage loop:

• ChatGPT Projects / Gemini Notebooks / NotebookLM Enterprise make project memory persistent.
• Glean / Microsoft Graph / SharePoint agents make enterprise memory permissioned and searchable.
• Notion Custom Agents make workspace data executable.
• Codex / Claude Code / Cursor / Copilot agents make codebases executable memory.
• mem0 / Letta make memory an explicit infrastructure layer.

Memova's opportunity is to own the missing personal layer:

natural inbox -> structured personal/project memory -> agent alignment -> confirmed action -> memory writeback.

Memova Framing

Old:

Inbox -> notes -> folders -> human remembers -> human synthesizes -> human outputs.

ChatGPT:

Inbox -> human selects context -> chat draft -> human moves output.

Agent + LLM wiki:

Inbox -> raw memory -> structured wiki -> agent alignment -> tool execution -> output -> memory writeback.

Memova's product wedge:

The future workflow is not "better note-taking" or "better chat." It is a closed loop where natural capture becomes structured memory, structured memory aligns agents, agents produce actions, and actions update memory.

Related

• Memova
• memova_agent_usage_chart_2026-05-23
• memova_product_layer_concept_2026-05-20