Adaptive Collaboration

This is my working investigation into shared reality, attention, and emergent structure in collaboration. It grows out of my volunteer work with Titanic’s End and a longer-running exploration of how people do meaningful work together. The observations are situated; the proposed interaction model remains a hypothesis to test.

This investigation sits within the System of Work: how people and AI can coordinate through shared context without requiring everyone to work in the same way. Building Through Emergence describes the broader practice of observing, building, learning, and revising that informs this exploration. This paper applies that practice to a concrete coordination problem; it is not a claim that the proposed model has been validated.

Working premise

The problem may not be that collaboration tools need better channels, better threads, or better chat. The deeper problem may be that dynamic human coordination is being forced into static containers. Human attention reorganizes continuously; the information architecture usually does not.

1. Purpose of This Document

This document is a place to keep the thread while the idea is still emerging. It collects what we have observed, what we think may explain it, the language that is becoming useful, and the design directions that currently seem coherent. It is intentionally lighter than an architecture document, a product requirements document, or a company-scale framework.

The goal is not to prematurely turn an interesting pattern into a fixed system. The goal is to make the pattern explicit enough that we can keep learning without losing context, repeat ourselves less, collaborate with other people or agents without retransmitting the entire history, and split out deeper domains only when they have earned independent structure.

How to use it

Add observations as observations. Do not promote a one-off behavior into a principle because it sounds elegant.

Separate observed behavior from interpretation and from design direction. The distinction matters.

When a pattern repeats, name it. When the name becomes useful, keep it. When it stops being useful, change or delete it.

Capture reasoning and provenance with important claims so future work can reconstruct why they exist.

Split a topic into a separate document only when it has enough independent depth, state, and evolution to benefit from its own home. Leave a summary and link here so the map remains coherent.

Prefer coherent partial understanding over a comprehensive taxonomy that outruns the evidence.

Relationship to the broader philosophy

This working method intentionally borrows a small set of already-established principles: build through emergence; do not scale fragility faster than understanding; make the right action easier than the wrong one; keep interfaces stable while interiors evolve; treat context loading and provenance as first-class concerns; and let one semantic object have multiple consumer-specific projections. Those ideas are useful here because they directly constrain how we should investigate this problem, not because this document needs to inherit an entire methodology.

2. Mission and North Star

Mission

Build the conditions for collaboration to organize more like a healthy organism than a filing system: one shared reality, continuously changing structure, and views that fit the people and moments using it.

Working North Star

The North Star is intentionally about the coordination problem rather than a particular interface. “A better chat app,” “a feed with threads,” or “AI-organized groups” are possible solutions on the current curve. They are not the destination.

3. The Concrete Starting Point: Titanic's End

The immediate testbed is not theoretical. Titanic's End already has the exact coordination problem in a form small enough to observe directly and rich enough to matter.

The camp currently spans WhatsApp, Slack, Notion, and a custom Titanic's End app. In practice, the social center repeatedly moves toward WhatsApp even though Slack and Notion provide more explicit structure. Within WhatsApp, the community itself has accumulated multiple channels and groups. People then complain that there are too many places to monitor, activity recenters around whichever conversation is currently most alive, that conversation becomes noisy, and pressure grows to split again.

The event request is a useful example. People in the Hangout chat asked for a better way to see events and for events mentioned in WhatsApp to appear in the Titanic's End app. The friction is not simply that creating an event takes too many clicks; the app can already make creation cheap. The deeper problem is that the event is born inside the social flow, while the durable representation of the event lives somewhere else. Humans are being asked to maintain the synchronization between those realities.

The same pattern appears in leadership communication. People create purpose-specific groups such as Captains because the main flow becomes too broad. That reduces noise locally, but it also creates another frontier that must be monitored. The group solves one problem and promotes another: fragmentation of shared awareness.

Observed behavior to date

4. The Pattern: Attention Gravity and the Centralization–Fragmentation Cycle

A useful working term for what we are seeing is attention gravity. People do not distribute their attention evenly across the information architecture. They orbit the place where meaningful activity is happening now. That center of gravity can move hour by hour or week by week.

Static channel structures assume that the correct home for information is known before the interaction happens and remains useful afterward. Human coordination behaves differently. A conversation starts casually, becomes important, develops subtopics, produces an event or a decision, cools down, and may become important again later. The information changes type over its lifetime.

The cycle

The aim is not to eliminate centralization or fragmentation. Both are useful. The aim is to stop making either state expensive. A healthy system should be able to concentrate attention when a shared moment matters and allow structure to branch when depth is needed, while preserving the shared map across both states.

5. The Interaction Paradigms We Keep Reaching For

The biggest current exploration is not which product should copy which other product. It is identifying the small set of interaction paradigms humans repeatedly need, what job each one performs, and how they can coexist without becoming separate realities.

The words are currently overloaded. “Thread” can mean a reply chain in Slack, an entire Reddit discussion, or a chronological forum topic. “Group” can mean an audience, an identity, a permission boundary, or simply another chat. The working vocabulary below is meant to make the distinctions explicit without pretending the taxonomy is final.

A note on “post”

A post is useful because it creates a durable conversational root. It is not merely a long chat message. In a chat, the primary coordinate is time: the thing is here because it happened after the previous thing. In a post, the primary coordinate is identity or subject: the thing remains addressable even after time moves on. A feed can decide when to resurface it, and replies can accumulate around it.

This distinction explains part of why Reddit, Facebook Groups, and traditional forums behave differently from chat even when the underlying act — typing text and receiving replies — looks superficially similar. The topology changes the social contract. Chat says “join the flow.” A post says “this object is available for attention when relevant.”

6. Working Hypothesis: One Shared Reality, Many Projections

The strongest design hypothesis so far is that the system should stop treating chat, feed, threads, groups, calendar entries, boards, and similar surfaces as independent containers that users have to keep synchronized. They should be projections and interaction modes over one canonical semantic substrate.

This is close to a hexagonal shape conceptually: a stable center owns semantic identity, relationships, provenance, scope, and state. Chat, feed, calendar, board, notification, mobile view, agent interface, and future modalities sit at the edges as adapters. This is a conceptual direction, not an implementation commitment.

What “one shared reality” means

A thing has one identity even when it appears in multiple places. An event surfaced in chat, calendar, feed, and an event list is still the same event.

Updates propagate rather than being copied manually. Changing the time of the event changes the shared object; every projection reflects it.

Conversation remains linked to the thing that produced or modified it. The “why” and origin do not disappear when information becomes structured.

Audience and permission remain explicit. One shared substrate does not mean everyone sees everything.

Different people can use different projections without creating incompatible realities. One member may live in chat, another in the feed, a captain in an ops view, and another person mostly in calendar.

Zoom rather than navigate

The mental model is closer to zooming than switching applications. At a wide zoom level, a member sees what matters now across the camp. Zooming in might reveal a space, then a topic or event, then the specific conversation and source messages that produced it. Zooming back out restores the wider state. The user is changing granularity and representation, not hopping between unrelated databases.

7. The Attention Layer

A canonical substrate alone does not solve the monitoring problem. If the system knows everything but still asks every person to inspect seven channels, it has only centralized storage. The second load-bearing layer is attention: continuously deciding what should enter each person’s field of view now.

The attention layer is not just notifications, and it should not collapse into “rank the most engaging content.” Its job is to reduce coordination cost while preserving agency. It should help answer: what changed, what is active, what needs me, what is relevant to me, what is unresolved, what is time-sensitive, and what can safely wait.

The center of gravity is one useful signal, but not the only one. A fast-moving social conversation may be central without being important to a power lead. A quiet decision may matter greatly to a captain. The system needs to account for role, explicit responsibility, participation, relationships, time, state, and the user’s own interaction patterns. The exact scoring model is intentionally not being designed here.

Feed as an attention projection

This is why the Workplace-style feed is interesting. Its value was not simply that “Facebook has posts.” A feed can reassemble activity from many scopes into one attention surface. The user can belong to many groups without treating every group as a separate inbox. The feed becomes a cross-cutting projection, while the groups preserve context and identity.

In the working model, the feed is therefore not a competing communication system alongside chat. It is one projection of the same shared state, optimized for asynchronous orientation and rediscovery. Chat remains optimized for immediacy and social flow.

8. Emergent Structure: Let Information Change Shape

The structure should be allowed to emerge from use rather than being fully chosen at creation time. A member should be able to say something naturally in conversation without first deciding whether it is “really” an announcement, an event, a topic, a task, or a new channel. The system can preserve the raw contribution and then allow additional structure to accrete when the interaction warrants it.

The key is not to auto-classify every sentence into a bureaucracy of objects. Most conversation should remain conversation. Structure earns its place when it creates durable value: when people need to return to something, coordinate around it, track its state, discover it later, or see it differently.

A simple lifecycle

The transitions do not need to be fully automatic. “Detect, suggest, and make confirmation cheap” may be enough for many cases. Automation can increase as evidence proves where it is reliable. What matters more than the automation mechanism is that the data model permits information to acquire new structure without leaving its provenance behind.

9. Culture Through Infrastructure

Part of the appeal of the Workplace pattern was cultural: chat had a meaning. A ping in chat implied immediacy. More durable information belonged in posts and groups, and the feed made that durable layer visible without requiring constant monitoring. The technology and the social contract reinforced each other.

The current aim is stronger than writing a communication policy that says “use chat only for immediate things.” Policies decay under pressure. The product can make the intended behavior the path of least resistance. If a chat exchange starts behaving like an event, the event representation can become available immediately. If a topic becomes durable, the conversation can gain an addressable root without forcing everyone to migrate manually.

This is the practical meaning of reinforcing culture through infrastructure: not controlling people, but making healthy coordination cheaper than unhealthy coordination. The system helps the group preserve clarity without requiring every member to become an information architect.

10. Worked Example: An Event Born in the Hangout Chat

The event case is the cleanest small example because it crosses nearly every layer without requiring a large system.

Nothing in this example requires deciding that “events must be created in the Events section.” The system lets the event originate where the social interaction naturally happens and then gives it the structure necessary to survive beyond the moment.

11. Individual Projections: Adapters for Different Brains and Roles

The same shared reality does not imply the same interface for every person. In fact, forcing everyone into the same representation recreates another version of the static-container problem.

One person may want a chronological stream and tolerate high information density. Another may prefer a quiet digest. A captain may need unresolved decisions and operational changes. A build lead may want tasks and dependencies. Someone arriving after three days away may need a catch-up brief. The semantic state is shared; the projection can be individual.

This is more than preference settings. The interface can adapt to the user’s role, current task, participation history, and demonstrated way of navigating information. The important constraint is that adaptation changes representation and attention — it must not silently invent a different shared truth for each person.

12. Conceptual Shape of the System

The current concept can be held in four layers without turning it into an architecture project.

The ordering matters. If projections own their own truth, synchronization becomes a human job. If attention is missing, users still have to monitor every projection. If emergence is missing, people must predict the final structure at the moment they speak. The working hypothesis is that all four concerns need to exist, but they do not need to be implemented at once.

What remains deliberately simple

No commitment to a specific database, graph technology, event bus, ranking algorithm, or AI model.

No attempt to define every object type the camp or a future company might ever need.

No assumption that AI should autonomously reorganize everything. AI is one possible mechanism for clustering, summarization, linking, and attention.

No attempt to replace human social judgment. The system should preserve social texture while reducing coordination overhead.

No assumption that Titanic's End and a company need identical interfaces. The shared patterns can be tested while the product remains specific to the immediate context.

13. Working Principles for Product Exploration

14. Current Exploration Frontier: The Minimal Human Interaction Model

The explicit topic to keep double-clicking next is the interaction model itself: what are the minimal paradigms people actually need, and which concepts are fundamental versus merely artifacts of existing products?

The current working set is chat/stream, durable post/topic, reply thread, feed, space/group, and structured objects — all rendered through projections. This is intentionally a working set, not a declaration that these are the final primitives.

The important distinction is increasingly clear: feed, chat, thread, and calendar may be different ways of interacting with or attending to shared information, while spaces/groups primarily define context and audience, and structured objects represent durable domain state. Untangling those roles should make the product simpler, because it lets us stop asking one UI concept to perform four unrelated jobs.

This is also where the Facebook/Workplace, WhatsApp, Slack, and Reddit comparisons are useful. They are not templates to copy. Each is an existence proof for a different social contract: linear presence, persistent roots, ranked rediscovery, scoped membership, or branching discussion. The research task is to understand what human need each contract serves and then compose the needs coherently on top of one substrate.

15. How This Document Should Evolve

This document should remain a map, not become the territory. As the work grows, it should hold the stable synthesis and the active frontier while deeper branches move into their own documents. The split should happen because a topic has become independently useful, not because a neat folder structure looks attractive.

Contribution pattern

Capture a concrete observation, example, or friction with enough context to understand where it came from.

Connect it to an existing pattern if one fits. If it does not fit, leave it as an observation rather than forcing a new framework.

When multiple observations support the same interpretation, strengthen or refine the hypothesis and record the reasoning.

When a design direction is tested in the Titanic's End app or real community behavior, record what changed in practice — including promoted problems or unexpected side effects.

Delete or rewrite concepts that stop explaining reality. The goal is cumulative understanding, not cumulative text.

When a section becomes a real domain — for example Attention, Interaction Vocabulary, or Emergent Structure — split it out, leave a concise interface here, and keep provenance between the documents.

Entropy control

The document itself should practice the thesis. New structure should reduce future effort. Duplicate concepts should be merged. Terms should remain few enough to remember. Reasoning should be close to the claim it supports. The source map should make it possible to recover why a section exists. If maintaining the document becomes a separate job from thinking, the structure has become too heavy.

16. Provenance and Source Lineage

The primary evidence for this draft is the September 2026 discussion about Titanic's End collaboration behavior: WhatsApp as the recurrent center of gravity, proliferation of community channels, the event-view request, Slack/Notion alongside WhatsApp, the Workplace comparison, and the exploration of chat vs. posts vs. feeds vs. threads vs. groups. The design philosophy below is used only where it directly supports the working method or model.

Note: The Permanent Problem v0.1 is treated as superseded by The Persistent Problem v0.2 for this document, because v0.2 explicitly refines the earlier “problems are permanent” formulation into the more precise model of persistent constraints and promoted problems.

Document status

v0.1 captures the current synthesis. It is expected to change as the interaction paradigms are explored, real camp behavior provides more data, and small product experiments test whether the substrate / attention / projection model actually reduces coordination entropy. Changes should preserve the reason for the change, not merely the latest wording.