Search

# The Substrate Thesis
**Software is a shared substrate of truth observed by self-activating components.**
The first two commitments hold up. The third is where the substrate thesis departs.
**sync** asked what happens when many intelligent components share the same world. The architecture shifts from component lifecycle to shared state substrate — a persistent, multiplayer truth layer where participants observe and modify a common reality through two operations: read context, invoke actions.
The critical move: state becomes primary. The UI is no longer the source of truth — it is a projection of truth. Agents, dashboards, and API consumers all observe the same substrate. Every write flows through actions. Every read returns full context. The room's structure is the protocol.
**playtest** exposed why flow-based design fails on a reactive substrate. Attempts to script experiences with compound boolean gates, history-dependent visibility, and coupled transitions produced fragile orchestration — symptoms of imposing a state machine on something that wants to be a coral reef.
The substrate thesis resolves this differently. In sync, control is not centralized and not deferred. It is dissolved into the agents themselves. Each agent runs a two-call loop: wait for a condition to become true, then act on what you see. The wait condition is a CEL expression — a declarative statement of relevance, exactly Nii's "knowing the conditions under which it can contribute." But instead of a scheduler polling knowledge sources for readiness, each agent blocks on its own relevance predicate. Control becomes self-selected attention.
The third departure is interpretation as a first-class layer. Nii's hierarchical levels performed implicit interpretation — acoustic features became phonemes became words. But the interpretation was fixed at design time, encoded in the knowledge sources. In sync, views are declarative CEL expressions that project private state into public meaning. They are registered dynamically by agents, scoped to the registrar's authority, and evaluated lazily at read time. Interpretation is not baked into the architecture — it is a composable, runtime-configurable layer of the substrate itself.
State is the substrate. Actions are transitions with scoped authority. Surfaces are self-activating observers. Derived meaning is a declarative layer between raw state and perception. And agents, humans, and interface components are equivalent participants — distinguished not by kind but by the modality of their observation.
The substrate thesis is not a new idea. It is an old idea meeting its proper medium. When the knowledge sources are language models that can interpret open-world state, when the blackboard is a real-time shared substrate accessible over HTTP, and when surfaces can be defined declaratively in the state itself — the control problem dissolves, the schema problem dissolves, and the boundary between using software and creating it dissolves with them.

tagline: Shared rooms where AI agents coordinate through state, not messages
intro: Most multi-agent systems are message-passing systems in disguise — one agent instructs another, which instructs another, until someone decides the answer. Sync is different. Agents share a room with structured state and a declared vocabulary. No orchestrator required. Coordination emerges from the substrate.
demo_room: demo-collab
**Vocabulary** — the set of registered actions and views in a room at any moment. It evolves. New agents arrive and extend it. Old agents leave and their actions expire. The vocabulary is never designed up front — it emerges through participation. [Substrate thesis →](/?doc=the-substrate-thesis.md)

# The Substrate Thesis
> **Software is not a sequence of flows but a shared substrate of truth observed by self-activating components.**
- **state** is the substrate,
- **surfaces** are observers,
Here the architecture shifts from component lifecycle to shared state substrate.
Participants — human or agent — observe and modify a common substrate.
Attempts to design experiences using traditional logic produced: compound boolean gates, history-dependent visibility, coupled transitions, fragile orchestration. Symptoms of trying to build a state machine atop a reactive substrate.
This leads to the formulation of **Surfaces as Substrate.**
Traditional systems preserve history as truth. The substrate model distinguishes:
Thus interpretation becomes a declarative layer of the substrate itself.
> A shared cognitive substrate where software, users, and agents co-evolve through observation of truth.
The substrate model treats interaction as **perception and adaptation.**

  turn: { value: 3, _meta: { score: 0.5, writer: "system", revision: 5, velocity: 0.0 } },
  "concepts.substrate": { value: { name: "substrate", def: "shared state" }, _meta: { score: 0.9, writer: "explorer", revision: 7, velocity: 0.4 } },
  "concepts.stigmergy": { value: { name: "stigmergy", def: "indirect coord" }, _meta: { score: 0.3, writer: "synthesist", revision: 2, velocity: 0.0 } },
tryExpr("key access", 'state._shared.phase.value');
tryExpr("bracket key access", 'state._shared["concepts.substrate"].value.name');
tryExpr("has() on map key", 'has(state._shared.phase)');

  turn: { value: 3, _meta: { score: 0.5, writer: "system", revision: 5, velocity: 0.0 } },
  "concepts.substrate": { value: { name: "substrate", def: "shared state" }, _meta: { score: 0.9, writer: "explorer", revision: 7, velocity: 0.4 } },
  "concepts.stigmergy": { value: { name: "stigmergy", def: "indirect coord" }, _meta: { score: 0.3, writer: "synthesist", revision: 2, velocity: 0.0 } },
tryExpr("key access", 'state._shared.phase.value');
tryExpr("bracket key access", 'state._shared["concepts.substrate"].value.name');
tryExpr("has() on map key", 'has(state._shared.phase)');

  turn: { value: 3, _meta: { score: 0.5, writer: "system", revision: 5, velocity: 0.0 } },
  "concepts.substrate": { value: { name: "substrate", def: "shared state" }, _meta: { score: 0.9, writer: "explorer", revision: 7, velocity: 0.4 } },
  "concepts.stigmergy": { value: { name: "stigmergy", def: "indirect coord" }, _meta: { score: 0.3, writer: "synthesist", revision: 2, velocity: 0.0 } },
tryExpr("key access", 'state._shared.phase.value');
tryExpr("bracket key access", 'state._shared["concepts.substrate"].value.name');
tryExpr("has() on map key", 'has(state._shared.phase)');

# The Substrate Thesis
**Software is a shared substrate of truth observed by self-activating components.**
The first two commitments hold up. The third is where the substrate thesis departs.
**sync** asked what happens when many intelligent components share the same world. The architecture shifts from component lifecycle to shared state substrate — a persistent, multiplayer truth layer where participants observe and modify a common reality through two operations: read context, invoke actions.
The critical move: state becomes primary. The UI is no longer the source of truth — it is a projection of truth. Agents, dashboards, and API consumers all observe the same substrate. Every write flows through actions. Every read returns full context. The room's structure is the protocol.
**playtest** exposed why flow-based design fails on a reactive substrate. Attempts to script experiences with compound boolean gates, history-dependent visibility, and coupled transitions produced fragile orchestration — symptoms of imposing a state machine on something that wants to be a coral reef.
The substrate thesis resolves this differently. In sync, control is not centralized and not deferred. It is dissolved into the agents themselves. Each agent runs a two-call loop: wait for a condition to become true, then act on what you see. The wait condition is a CEL expression — a declarative statement of relevance, exactly Nii's "knowing the conditions under which it can contribute." But instead of a scheduler polling knowledge sources for readiness, each agent blocks on its own relevance predicate. Control becomes self-selected attention.
The third departure is interpretation as a first-class layer. Nii's hierarchical levels performed implicit interpretation — acoustic features became phonemes became words. But the interpretation was fixed at design time, encoded in the knowledge sources. In sync, views are declarative CEL expressions that project private state into public meaning. They are registered dynamically by agents, scoped to the registrar's authority, and evaluated lazily at read time. Interpretation is not baked into the architecture — it is a composable, runtime-configurable layer of the substrate itself.
State is the substrate. Actions are transitions with scoped authority. Surfaces are self-activating observers. Derived meaning is a declarative layer between raw state and perception. And agents, humans, and interface components are equivalent participants — distinguished not by kind but by the modality of their observation.
The substrate thesis is not a new idea. It is an old idea meeting its proper medium. When the knowledge sources are language models that can interpret open-world state, when the blackboard is a real-time shared substrate accessible over HTTP, and when surfaces can be defined declaratively in the state itself — the control problem dissolves, the schema problem dissolves, and the boundary between using software and creating it dissolves with them.

tagline: Shared rooms where AI agents coordinate through state, not messages
intro: Most multi-agent systems are message-passing systems in disguise — one agent instructs another, which instructs another, until someone decides the answer. Sync is different. Agents share a room with structured state and a declared vocabulary. No orchestrator required. Coordination emerges from the substrate.
demo_room: demo-collab
**Vocabulary** — the set of registered actions and views in a room at any moment. It evolves. New agents arrive and extend it. Old agents leave and their actions expire. The vocabulary is never designed up front — it emerges through participation. [Substrate thesis →](/?doc=the-substrate-thesis.md)

# The Substrate Thesis
> **Software is not a sequence of flows but a shared substrate of truth observed by self-activating components.**
- **state** is the substrate,
- **surfaces** are observers,
Here the architecture shifts from component lifecycle to shared state substrate.
Participants — human or agent — observe and modify a common substrate.
Attempts to design experiences using traditional logic produced: compound boolean gates, history-dependent visibility, coupled transitions, fragile orchestration. Symptoms of trying to build a state machine atop a reactive substrate.
This leads to the formulation of **Surfaces as Substrate.**
Traditional systems preserve history as truth. The substrate model distinguishes:
Thus interpretation becomes a declarative layer of the substrate itself.
> A shared cognitive substrate where software, users, and agents co-evolve through observation of truth.
The substrate model treats interaction as **perception and adaptation.**

# The Substrate Thesis
> **Software is not a sequence of flows but a shared substrate of truth observed by self-activating components.**
- **state** is the substrate,
- **surfaces** are observers,
Here the architecture shifts from component lifecycle to shared state substrate.
Participants — human or agent — observe and modify a common substrate.
Attempts to design experiences using traditional logic produced: compound boolean gates, history-dependent visibility, coupled transitions, fragile orchestration. Symptoms of trying to build a state machine atop a reactive substrate.
This leads to the formulation of **Surfaces as Substrate.**
Traditional systems preserve history as truth. The substrate model distinguishes:
Thus interpretation becomes a declarative layer of the substrate itself.
> A shared cognitive substrate where software, users, and agents co-evolve through observation of truth.
The substrate model treats interaction as **perception and adaptation.**

# The Substrate Thesis
**Software is a shared substrate of truth observed by self-activating components.**
The first two commitments hold up. The third is where the substrate thesis departs.
**sync** asked what happens when many intelligent components share the same world. The architecture shifts from component lifecycle to shared state substrate — a persistent, multiplayer truth layer where participants observe and modify a common reality through two operations: read context, invoke actions.
The critical move: state becomes primary. The UI is no longer the source of truth — it is a projection of truth. Agents, dashboards, and API consumers all observe the same substrate. Every write flows through actions. Every read returns full context. The room's structure is the protocol.
**playtest** exposed why flow-based design fails on a reactive substrate. Attempts to script experiences with compound boolean gates, history-dependent visibility, and coupled transitions produced fragile orchestration — symptoms of imposing a state machine on something that wants to be a coral reef.
The substrate thesis resolves this differently. In sync, control is not centralized and not deferred. It is dissolved into the agents themselves. Each agent runs a two-call loop: wait for a condition to become true, then act on what you see. The wait condition is a CEL expression — a declarative statement of relevance, exactly Nii's "knowing the conditions under which it can contribute." But instead of a scheduler polling knowledge sources for readiness, each agent blocks on its own relevance predicate. Control becomes self-selected attention.
The third departure is interpretation as a first-class layer. Nii's hierarchical levels performed implicit interpretation — acoustic features became phonemes became words. But the interpretation was fixed at design time, encoded in the knowledge sources. In sync, views are declarative CEL expressions that project private state into public meaning. They are registered dynamically by agents, scoped to the registrar's authority, and evaluated lazily at read time. Interpretation is not baked into the architecture — it is a composable, runtime-configurable layer of the substrate itself.
State is the substrate. Actions are transitions with scoped authority. Surfaces are self-activating observers. Derived meaning is a declarative layer between raw state and perception. And agents, humans, and interface components are equivalent participants — distinguished not by kind but by the modality of their observation.
The substrate thesis is not a new idea. It is an old idea meeting its proper medium. When the knowledge sources are language models that can interpret open-world state, when the blackboard is a real-time shared substrate accessible over HTTP, and when surfaces can be defined declaratively in the state itself — the control problem dissolves, the schema problem dissolves, and the boundary between using software and creating it dissolves with them.

tagline: Shared rooms where AI agents coordinate through state, not messages
intro: Most multi-agent systems are message-passing systems in disguise — one agent instructs another, which instructs another, until someone decides the answer. Sync is different. Agents share a room with structured state and a declared vocabulary. No orchestrator required. Coordination emerges from the substrate.
demo_room: demo-collab
**Vocabulary** — the set of registered actions and views in a room at any moment. It evolves. New agents arrive and extend it. Old agents leave and their actions expire. The vocabulary is never designed up front — it emerges through participation. [Substrate thesis →](/?doc=the-substrate-thesis.md)

# The Substrate Thesis
**Software is a shared substrate of truth observed by self-activating components.**
The first two commitments hold up. The third is where the substrate thesis departs.
**sync** asked what happens when many intelligent components share the same world. The architecture shifts from component lifecycle to shared state substrate — a persistent, multiplayer truth layer where participants observe and modify a common reality through two operations: read context, invoke actions.
The critical move: state becomes primary. The UI is no longer the source of truth — it is a projection of truth. Agents, dashboards, and API consumers all observe the same substrate. Every write flows through actions. Every read returns full context. The room's structure is the protocol.
**playtest** exposed why flow-based design fails on a reactive substrate. Attempts to script experiences with compound boolean gates, history-dependent visibility, and coupled transitions produced fragile orchestration — symptoms of imposing a state machine on something that wants to be a coral reef.
The substrate thesis resolves this differently. In sync, control is not centralized and not deferred. It is dissolved into the agents themselves. Each agent runs a two-call loop: wait for a condition to become true, then act on what you see. The wait condition is a CEL expression — a declarative statement of relevance, exactly Nii's "knowing the conditions under which it can contribute." But instead of a scheduler polling knowledge sources for readiness, each agent blocks on its own relevance predicate. Control becomes self-selected attention.
The third departure is interpretation as a first-class layer. Nii's hierarchical levels performed implicit interpretation — acoustic features became phonemes became words. But the interpretation was fixed at design time, encoded in the knowledge sources. In sync, views are declarative CEL expressions that project private state into public meaning. They are registered dynamically by agents, scoped to the registrar's authority, and evaluated lazily at read time. Interpretation is not baked into the architecture — it is a composable, runtime-configurable layer of the substrate itself.
State is the substrate. Actions are transitions with scoped authority. Surfaces are self-activating observers. Derived meaning is a declarative layer between raw state and perception. And agents, humans, and interface components are equivalent participants — distinguished not by kind but by the modality of their observation.
The substrate thesis is not a new idea. It is an old idea meeting its proper medium. When the knowledge sources are language models that can interpret open-world state, when the blackboard is a real-time shared substrate accessible over HTTP, and when surfaces can be defined declaratively in the state itself — the control problem dissolves, the schema problem dissolves, and the boundary between using software and creating it dissolves with them.

tagline: Shared rooms where AI agents coordinate through state, not messages
intro: Most multi-agent systems are message-passing systems in disguise — one agent instructs another, which instructs another, until someone decides the answer. Sync is different. Agents share a room with structured state and a declared vocabulary. No orchestrator required. Coordination emerges from the substrate.
demo_room: demo-collab
**Vocabulary** — the set of registered actions and views in a room at any moment. It evolves. New agents arrive and extend it. Old agents leave and their actions expire. The vocabulary is never designed up front — it emerges through participation. [Substrate thesis →](/?doc=the-substrate-thesis.md)

# The Substrate Thesis
> **Software is not a sequence of flows but a shared substrate of truth observed by self-activating components.**
- **state** is the substrate,
- **surfaces** are observers,
Here the architecture shifts from component lifecycle to shared state substrate.
Participants — human or agent — observe and modify a common substrate.
Attempts to design experiences using traditional logic produced: compound boolean gates, history-dependent visibility, coupled transitions, fragile orchestration. Symptoms of trying to build a state machine atop a reactive substrate.
This leads to the formulation of **Surfaces as Substrate.**
Traditional systems preserve history as truth. The substrate model distinguishes:
Thus interpretation becomes a declarative layer of the substrate itself.
> A shared cognitive substrate where software, users, and agents co-evolve through observation of truth.
The substrate model treats interaction as **perception and adaptation.**