Patterns that needed a name.

First named · Apr 2026

Agentic Envoy

A diplomatic layer between autonomous agents that carries organizational intent as a first-class primitive. Current agent frameworks assume agents will cooperate. None of them specify how agents carry organizational context, exercise judgment about whether an interaction should happen, or stay accountable when acting on behalf of an organization. An envoy isn't a proxy. A proxy forwards blindly. An envoy understands who it represents, what the mission is, and has the authority to say "this doesn't align" in situations nobody anticipated in advance. Agent-to-agent communication becomes envoy-to-envoy negotiation. The envoy is the observation layer, the security layer, and the access control layer, all at once.

First named · Mar 2026

Agent Bus

The message bus is a 40-year-old enterprise pattern. The agent bus is what happens when the consumers on that bus are autonomous, concurrent, and generative. Traditional message buses assume deterministic consumers, predictable throughput, and bounded output. An agent bus has non-deterministic consumers, variable latency, and every consumer can generate unbounded output in response to a single message. This is fundamentally different from A2A protocols, which are point-to-point conversations. The agent bus is broadcast by nature: n:m topology where multiple publishers and multiple consumers operate simultaneously. Most current agent coordination breaks at this boundary because nobody has named the topology or built primitives native to it.

First named · Mar 2026

Speedtrap

A low-tech output guardrail for agent bus systems. Before an agent posts a response, it must prove it read the latest message in the channel. If new messages arrived while the agent was processing, those messages get fed back into its context, and the agent decides whether its response is still relevant. The name captures what it does: it catches agents moving too fast to look at what's ahead of them. The first coordination primitive designed natively for n:m agent bus topologies.

First named · Feb 2026

Compaction Amnesia

The gradual loss of meaningful context when AI memory systems compress or summarize conversational state. The standard response is better audit trails and recovery scripts. The deeper insight: compaction amnesia is a symptom, not the disease. The disease is storing memory as raw state that a summarizer can butcher. If memory is already stored as compressed reasoning trajectories with structured frontmatter, compaction doesn't destroy meaningful state because the meaningful structure was never stored as raw context in the first place.

First named · Dec 2025

The Three-Layer AI Taxonomy

A classification of AI use cases that reveals where the real frontier is. Layer 1: Do Stuff. Personal productivity, task execution, automation. This is where 95% of the AI conversation lives. Layer 2: Think Stuff. AI as a thinking partner, generating reframes, perspectives, connections. The unmapped frontier. Layer 3: Build Stuff. Commercial materialization, organizational agents, products. Layer 2 is massively underrated because it produces no visible artifact. Nobody screenshots a thinking session. The value is upstream of everything visible.

First named · Nov 2025

Constrained Combination

A theory of what makes thinking different from pattern matching. LLMs have all the knowledge ever written down by humanity but none of the experience of applying it. The common framing says LLMs lack experience. The sharper framing: they lack constraint. Thinking isn't combination. Thinking is constrained combination. The constraint is the cognition. Your specific cognitive architecture, your experience, your wiring determines which combinations are possible and which are invisible. LLMs have knowledge without constraint. Data without substrate.

First named · Nov 2025

The Speech Center Model

A cognitive architecture model for understanding what LLMs actually are. The LLM isn't the brain. It's the speech center of the brain. Foundation docs function as the limbic system (authoritative but invisible). The vault and working knowledge function as the neocortex. Edges and memory paths function as the hippocampus. The LLM itself is Broca's and Wernicke's areas: language rendering, not cognition. The associative pattern-making happens elsewhere. This reframes every conversation about "AI thinking" and explains why memory architecture matters more than model size.

First named · Mar 2026

Conversational First

A design philosophy where the primary interface is a conversation and visual elements exist to support that conversation, not the other way around. Distinct from AI-first, which bolts chat onto existing UI. Conversational First inverts the relationship: the conversation is the product, the screen is the scaffold. When the interface becomes a conversation, the product surface changes. So does pricing. So does what "a feature" even means.

Read the essay  →

First named · Dec 2025

Liberating Infrastructure

The discipline of designing systems whose success is measured by how invisible they become and how much agency they return to what flows through them. Whisperline senses so organizations can be what they actually are. CCA structures voice so content can express authentically. ReflectiveAI structures memory so thinking can flow. Intent-Driven Architecture structures implementation so users can simply want. The pattern across all of them: create architecture that returns sovereignty to what flows through it. Structure that wants to disappear.

First named · Nov 2025

Intent-Driven Architecture

Start from intent, not implementation. Most software makes users adapt to someone else's workflow. Intent-Driven Architecture inverts that: the user states what they're trying to accomplish, and the system figures out how to serve it. The UI doesn't disappear. It becomes negotiable.

First named · Oct 2025

Content Creation Architecture

Separation of concerns applied to content creation. Most brand guidelines fail because they mix five concerns into one document: what's always true, what this piece needs, how it should sound, how it gets made, how it gets checked. Changing tone breaks vocabulary. Adding examples changes structure. Everything is coupled. CCA treats this as an engineering problem, not a creative one. The layers change independently, combine freely, and work with any generation method, human or AI.

First named · Feb 2026

Context Rot

The gradual decay of shared understanding when context stops being actively maintained. Conversations start from zero. Decisions lose their rationale. Institutional knowledge erodes silently. Different from information loss: context rot happens even when the documents still exist, because the reasoning behind them has disconnected from the people who need it.

First named · Dec 2025

The Categorization Tax

The cost of premature categorization. Every time a tool demands you classify something before you understand it, you pay a tax. "Is this a task or an issue? A user story or an epic? P2 or P3?" Hours spent serving the tool's data model instead of capturing the actual insight. Fifty years of enterprise software has been constrained by relational database thinking: schema before data. The categorization gate is where insight goes to die. With vector stores, graph databases, and LLMs, the technological constraint is gone. The mental model persists.

First named · Oct 2025

The Two-Worlds Framework

Every organization operates in two realities simultaneously. The formal world is what gets measured: dashboards, org charts, quarterly reviews. The informal world is what gets felt: trust, tension, fatigue, who actually influences decisions, what people notice but don't say. Most leaders manage the formal world while the informal world quietly shapes everything. By the time formal metrics shift, the informal reality moved weeks ago. The gap between these two worlds is where organizational risk hides. And opportunity.

First named · Jan 2025

EAOS

When emotion becomes a navigational signal, numbers regain meaning. EAOS (Emotion-Aware Organizational Sensing) is an interpretive layer that turns anonymous emotional signals into structured organizational insight. Not a survey. Not sentiment analysis. A sensing system that captures what people actually feel, not what they report feeling, and translates that into patterns leaders can act on.