Tinderbox Meetup 04Oct26: Learning How to Say What We Don’t Yet Know How to Say

Training Videos
1

Tinderbox Meetup 04Oct26: Learning How to Say What We Don’t Yet Know How to Say
Level Intermediate
Published Date 10/7/26
Type Meetup
Tags AI-Augmented Thinking, Artificial Intelligence, Human Agency, Incremental Formalization, Knowledge Management, Tools for Thought, 5CKM, 5Cs of Knowledge Management, Eastgate, Identity Praxis, Inc., Mark Bernsetein, Michael Becker, Tinderbox
Video Length 01:40:04
Video URL https://youtu.be/Io5smQL0Ee8
Chat File TBX Meetup 04Oct26_chat.txt (17.3 KB)
TBX Version 11.5
Host Michael Becker

This meetup explored the relationship between Tinderbox, tools for thought, AI, reading, writing, knowledge management, and software development. Mark Bernstein opened with reflections from the Hypertext conference, including work with Andreas Grimm and the provocative idea that the purpose of tools for thought is to help us “find out how to say what we don’t yet know how to say.” This became an important theme throughout the discussion: tools are valuable not simply because they make us more productive, but because they can help us articulate, discover, and develop thoughts that are not yet fully formed.

The conversation then moved into AI-assisted reading and writing, human agency, anthropomorphism, summaries versus understanding, and the role of AI in helping people articulate their thinking. Michael Becker demonstrated an evolving knowledge collection and curation system designed around intentional reading, structured capture, AI, and Tinderbox. This led to a broader discussion of vibe coding versus structured AI-leveraged development, documentation, privacy, security, modularity, and the importance of incremental formalization—allowing structure to emerge as understanding develops rather than imposing it prematurely.

Key Moments

  • 00:01:49 — Reflections from the Hypertext conference. Mark Bernstein describes the recent conference as one of the strongest Hypertext conferences in years and discusses renewed interest in tools for reading, writing, and thinking.
  • 00:03:03 — Obsidian, AI, MCP, and Tinderbox. Discussion of work examining an AI-assisted writing process using an Obsidian vault and MCP, and how similar approaches could be implemented in Tinderbox.
  • 00:04:53 — Are productivity metrics appropriate for tools for thought? Mark questions whether output-per-hour is the right way to evaluate thinking tools. The objective may instead be to produce changes in ourselves and our understanding.
  • 00:06:13 — Learning how to say what we don’t yet know how to say. Mark discusses his work with Andreas Grimm and offers a central idea for the meetup: the purpose of these tools is to help us discover how to articulate things that we do not yet know how to express.
  • 00:08:11 — MCP and reading alongside AI. Mark reflects on MCP and the growing willingness of people to incorporate large language models into reading and knowledge work.
  • 00:09:05 — “Don’t be jealous, just be you.” Michael discusses writing alongside Claude and asks how we maintain our own identity and contribution when AI can perform some tasks faster or better than we can.
  • 00:10:33 — Authenticity, voice, and “AI slop.” Where is the boundary between someone’s own voice and AI participation? What does it mean to have an authentic voice when working collaboratively with AI?
  • 00:12:02 — Should we anthropomorphize AI? Chuck argues against treating AI as a person and suggests thinking of it explicitly as a machine assistant.
  • 00:17:30 — AI can force us to articulate our thinking. Mark Anderson observes that explaining to an AI why it is wrong can force us to clarify ideas that previously existed only vaguely in our heads.
  • 00:22:35 — What gets lost when AI does the reading? The group discusses the danger of relying on AI summaries. An important insight buried deep within a long paper may disappear entirely from a summary.
  • 00:26:16 — “I am the agent of my AI.” Chuck emphasizes retaining human agency rather than thinking of AI as an autonomous actor directing the human.
  • 00:32:34 — Summary is not understanding. Mark Anderson discusses the limits of abstracts, keywords, and AI summaries. The idea that matters most to a reader may not be what the author—or an AI—considers the paper’s primary point.
  • 00:35:13 — Documentation, software, and institutional knowledge. Code can tell us what a system does without necessarily explaining why it was designed that way.
  • 00:38:46 — “Rendering the whale”: reading with intention. Michael demonstrates his developing knowledge-capture system. Instead of simply saving an article, the reader breaks it apart into useful elements—terms, people, places, definitions, quotations, arguments, notes, and actions.
  • 00:47:40 — Capturing intention and destination. Captured knowledge can include the reader’s intention, project, notes, metadata, and destination before being sent into systems such as Tinderbox.
  • 00:49:13 — AI-generated documentation of the development process. Michael shows reports documenting features, architecture, development decisions, and the evolution of the project.
  • 00:50:04 — Incremental formalization in practice. Features, refinements, and bug fixes reveal how a system evolves through many small iterations rather than through a completely predetermined specification.
  • 00:51:21 — Security, privacy, architecture, and data flows. The discussion turns to documenting the system’s architecture and actively testing privacy and security assumptions.
  • 00:58:47 — Vibe coding vs. structured leveraged AI. Michael distinguishes one-off AI-generated code from a disciplined process involving Git, documentation, security, privacy, modularity, intentional design, maintenance, and reversibility.
  • 01:01:40 — Automate what you repeatedly do. A lesson from the Tinderbox community: when you find yourself performing the same operation repeatedly, stop and ask whether the process itself can be automated.
  • 01:12:24 — The danger of premature formalization. Michael connects the discussion to Shipman & Marshall’s work and argues against imposing rigid structure before we sufficiently understand the problem.
  • 01:12:55 — “Embrace incremental formalization.” Rather than trying to specify everything in advance, progressively formalize ideas and systems as understanding develops—while maintaining enough modularity and structure to make later changes surgical.
  • 01:14:02 — Back to Tinderbox. The meetup returns to practical Tinderbox demonstrations and small capabilities that can make everyday knowledge work easier.

Central Idea

A strong thread connecting the entire meetup is:

Thought → Articulation → Structure

We often begin with something we sense or understand incompletely but cannot yet articulate. Tools for thought—and increasingly AI—can help us work with that ambiguity until we discover how to express it. Only then should we progressively impose structure.

In this sense, Mark Bernstein and Andreas Grimm’s idea that these systems help us “find out how to say what we don’t yet know how to say” connects directly with incremental formalization. The goal is neither to formalize our thinking too early nor to outsource the thinking to AI. Instead, tools should help us develop, articulate, organize, and refine our own understanding while preserving human agency.

Key Themes

Tinderbox; Tools for Thought; Hypertext; Artificial Intelligence; Human Agency; Knowledge Management; Personal Knowledge Management; Intentional Reading; AI-Assisted Reading; AI-Assisted Development; Incremental Formalization; Premature Formalization; Vibe Coding; Structured Leveraged AI; Documentation; Privacy; Security