Start here
Three ways into the ideas and practice behind this blog.
Latest writing
Modeling Thought Through Language: EchoPath Labs Is My First Step
My long-term goal is to model how people think through the words they express. EchoPath Labs is my first step, connecting meaning, bounded action, and cognitive history in engineering collaboration so that interpretations can be tested and revised through practice.
Why Multiple Agents, Rather Than One Universal Harness
A model fitting into a harness does not mean the resulting agent is trustworthy. This essay explains why a calibrated Harness + Model should be treated as a complete Agent Profile, and how RelayPact enables a trusted Host Agent to delegate bounded work to other agents.
Rejection Is Easy; Reconstruction Is Hard
Rejecting an old structure creates direction, enemies, and consensus. Real reconstruction requires understanding why it formed, what it still does, and which deeper conditions keep reproducing its failures.
Seeing the World Through Structuralism: Before Judging People, Look at the Structure
Why do organizations keep selecting the same kinds of people even after leaders change? A practical structuralist lens reveals the systems shaping behavior.
Agents Don't Need Org Charts; They Need Context Boundaries
Multi-agent systems may need clear boundaries around knowledge, state, permissions, tools, and decisions more than human-style job titles. Agents are replaceable; context is durable.
Software Organizations in the AI Era: From Multi-Person Teams to One-Person Organizations
AI may reshape software development less by accelerating coding than by changing how work is organized. Explore the shift toward human-governed, agent-executed teams.
The Stronger the Agent, the More We Need Software Engineering
AI makes code cheaper, not complexity free. Architecture, modularity, schemas, tests, and domain models reduce agent uncertainty through cognitive reuse.
Rethinking Password Management in the Agent Era
As AI agents act across GitHub, APIs, and cloud platforms, KeptNear's Local Broker shows how to authorize credential-backed capabilities without exposing raw secrets.
When AI Begins to Think Along Your Lines
When ChatGPT does more than remember preferences and begins extending a person's structure of thought, the central challenge of human-AI collaboration shifts from storing more information to sustaining a cognitive subject that is still changing.
Why I Started Simplifying My AI Agents
After intensive use of AI agents such as Codex and Claude Code, I began removing unnecessary skills, rules, and context. This article explains why more configuration is not always better and how restraint can make agents more reliable.