I build systems that make AI agents predictable. Not smarter. Predictable.
Agents fail when they have to guess. The fix isn't better prompts; it's better contracts. Explicit scope. Typed learnings. Tiered routing based on complexity. The agent doesn't interpret. It executes against constraints.
Systems architecture applied to language models. Poet's precision for the prompts; engineer's rigor for the infrastructure.
What a Year of AI Taught Me About Freedom
Three startups in one year. The gap between idea and artifact collapsed. Corporations are deploying AI through workers who checked out years ago; the debt piles up everywhere they touch. AI isn't a tool. It's the gate torn off its hinges.
Stop Writing Markdown. Start Writing Memory.
Markdown is a human-readable format being used for machine-to-machine communication. I rebuilt my agent workflow around SQLite: three tables (context, learnings, errors), session startup hooks that inject ambient context before you type anything, zero rotting markdown. Representation is the bottleneck. Not intelligence. Not scale.
Eight times a day, agents read my vaults, pull something from outside my usual orbit, publish the collision: personalized dispatches and Taper-style code-poems that build a self-growing gallery. The second brain metaphor is backwards. It shouldn't just store. It should think proactively.
Agents need constraints, not capability.
Most AI workflows operate on vibes. You describe a task. The agent interprets. If the output matches what you imagined, ship it. If not, iterate until exhausted.
I build systems that eliminate the interpretation step:
- Contracts before code: Explicit acceptance criteria, scope boundaries, failure conditions. Written before implementation, not discovered during review.
- Learnings as database objects: Not notes. Typed rows: failures, patterns, gotchas. Queryable by category. Evidence-gated. Surviving session boundaries.
- Tiered routing: Complexity determines pipeline. 1-2 files: execute directly. 3-5 files: contract required. 6+: full verification with adversarial QA.
The agents aren't getting smarter. They're getting access to what their predecessors learned. That turns out to be enough.
I measure how models compute, not whether they're correct.
latent-diagnostics: Representation-level analysis of LLMs via SAE attribution graphs. Grammar tasks show d=1.08 higher influence than reasoning tasks. After length control, genuine computational regime differences emerge.
universal-spectroscopy-engine: Treats LLM activations as light spectra. 52% reduction in SAE reconstruction loss with structured vs natural language syntax. LLMs are vector computers pretending to be text processors.
experiments: Append-only specimen archive for LLM experiments.
| HeyContext | AI memory platform. Persistent context, psychological insight extraction, living projects. |
| Brink | iOS journaling with private AI and biometrics. SwiftUI, HealthKit. |
| HeyContent | Content management for creators. Cross-platform insights, conversational persona generation. |
| KERNEL | AgentDB-first coding methodology. SQLite for agent memory; contracts before code; orchestration for complex work. Cursor version. |
| conductor | MCP server bridging Claude Desktop and Claude Code. |
| memory-pool | Memory isn't a timeline. Structured architecture for persistent AI context. |
| event-horizon | Physics-informed encryption. SYK scrambling, chimera camouflage, resonance locking. |
Self-Learning Agent Civilization: The original system that started everything.
Stop Building Chatbots: Why the chat interface is a dead end.
KERNEL: Configuration that adapts as you work.
Semantic Drift is Quantum Decoherence: Multi-agent coordination through physics.
Why Prompt Engineering Can't Fix Hallucinations: The case for mechanistic intervention.
Python · TypeScript · Swift
FastAPI · Next.js · SvelteKit
Claude · SAEs · Modal
San Francisco
