[ Blog_Posts ]

Determinism Where You Can, Judgement Where You Must: The Technique Boundary for AI Systems

Every part of an AI system has exactly one right home: a rule, a state machine, or a bounded judge. Control belongs to the orchestrator, not to an autonomous loop. A first-principles boundary for deciding which technique owns which part.

2026-07-09 Harrison Guo 18 min read generative-ai system-design reliability llm-as-judge rules-engine state-machine ai-agents neuro-symbolic first-principles verification evaluation

Shrink the Stochastic Surface: A Design Standard for Probabilistic Systems

The reliability of a probabilistic system is set by how much of it you let the model decide. A first-principles standard for shrinking the stochastic surface, on the generation side and the evaluation side.

2026-06-24 Harrison Guo 13 min read generative-ai system-design reliability retrieval-augmented-generation fine-tuning evaluation llm-as-judge rules-engine neuro-symbolic first-principles verification

Generative AI Builds Shapes, Not Games: The Constraint Gap and the Architecture That Closes It

Generative AI nails plausible shapes but misses constraints, composition, and verification. A first-principles read of the gap, and the architecture that closes it.

2026-06-20 Harrison Guo 12 min read generative-ai text-to-3d game-development procedural-generation neuro-symbolic constraint-satisfaction minecraft higgsfield ai-limitations first-principles verification

I Tested Higgsfield's Minecraft 'Prompt-to-Build.' It Generates Shapes, Not Scenes.

I ran eight building prompts through Higgsfield's Minecraft prompt-to-build. It nails single shapes in a minute but drops exact sizes, materials, doors, and whole scenes.

2026-06-18 Harrison Guo 11 min read higgsfield minecraft ai-tool-evaluation generative-3d text-to-3d voxel prompt-to-build ai-benchmark hands-on-review

Agent Architecture Is a Compute Allocation Problem: The Advisor Strategy, Cost-Curve Frame Recursed

Anthropic named the advisor strategy in April. Tobi Lutke made it viral in May with Qwen plus GPT-5.5. Stanford's HazyResearch formalized the same shape earlier. One cost-curve frame unifies all three: a cheap executor runs the loop, an expensive advisor weighs in only at hard decisions. The third recursion.

2026-06-15 Harrison Guo 21 min read claude-code ai-agent advisor-strategy anthropic tobi-lutke qwen cost-curve model-orchestration first-principles distributed-systems agent-architecture

Agent Retrieval Above the Crossover: A First-Principles Read of CodeGraph

CodeGraph is the LLM-symbol-graph my prior retrieval post argued should exist. Read against its own SQLite index: why its architectural choices are right, and where the abstraction leaks.

2026-06-08 Harrison Guo 25 min read claude-code codegraph ai-agent retrieval symbol-graph tree-sitter mcp sqlite fts5 first-principles distributed-systems abstraction-leaks

I Tested CodeGraph on Hono. The Tool-Call Savings Reproduce — the Cost Savings Don't.

Independent CodeGraph benchmark on Hono (~280 TS files): -55% tool calls reproduces the published claim, but cost is a wash (+7%), not -35%. Raw CSV included.

2026-06-01 Harrison Guo 19 min read claude-code codegraph ai-agent retrieval benchmark hono mcp tree-sitter claude-code-cost ai-tool-evaluation cost-curve

Agent Memory Is a Cache Coherence Problem

Lossless curated notes vs lossy auto-compression with vector recall: two AI-memory designs that fail differently. One fails like a cache — classical systems already mapped it.

2026-05-28 Harrison Guo 20 min read ai-agent agent-memory claude-mem claude-code distributed-systems cache-coherence rag vector-memory source-leak first-principles

Agent Retrieval Is a Cost Curve Problem: Why Claude Code Doesn't Use RAG

Why Claude Code uses grep instead of RAG: a cost-curve argument, audited against the source — with the Explore vs Fork A/B test nobody mentions.

2026-05-25 Harrison Guo 27 min read claude-code ai-agent rag retrieval distributed-systems cost-curve grep ripgrep source-leak explore-agent fork-subagent first-principles

Channels Aren't Message Passing — How Parked Goroutines OOM-Killed a Pod

Go channels are synchronization primitives, not queues. They deliver backpressure only when the producer is bounded — and that's where the real OOM hides.

2026-05-12 Harrison Guo 18 min read go channels concurrency goroutines oom backpressure production

[ Blog_Posts ]

[ Connect_With_Me ]