Blog
Essays on AI agents, retrieval, extraction, and evals — what holds up in production and what only works in a demo.
Everything Is Computer
Harbor turns jobs into containers, instructions, and graders — the bottleneck is the test.
Every Max Coding Agent Plan, Tested
Four $200 coding agents on the same work. Which surface makes it disappear?
Python to TypeScript in One Evening
Parallel agents and fast inference made the port possible. Deciding what exists stayed hard.
Build AI That Removes Work, Not Adds It
Untrusted AI creates more work. Measure total efficiency, not capability.
When Code Is Easy, Judgment Is Scarce
When AI makes code cheap, context, review, and judgment become scarce.
The New Programming Layer: Learning to Operate Alien Technology
Programming is getting a new layer: agents, prompts, MCP, and tools.
Do the Simplest Thing: Agent Design for Better Models
Keep agents simple and observable so stronger models can take over later.
Why Fast, Small Models Are the Way to Go
LLMs fail. Fast, graceful failures beat slow ones every time.
Why Open-Weight Models Matter for AI Independence
Open-weight models protect builders from bans, rate limits, and policy shifts.
The Anti-Slop Backlash: Keep Thinking in a Synthetic Sludge Age
Resist generative slop. Language-first tools still sharpen judgment.
Why Agentic Search is Replacing Traditional RAG
Agentic search replaces static RAG with iterative, intent-driven retrieval.
Why AI Agents Signal the End of Click-Based Computing
Agents replace click-based UIs with conversational, intent-based interaction.
How Large Language Models Think and Learn
How LLMs process, reason, and learn — context, in-context learning, and architecture.
The Next Generation of AI Coding Assistants
Claude Opus 4 and Sonnet 4: coding, extended thinking with tools, and Claude Code.
Connecting LLMs to Your Data
MCP resources let LLMs read external data while the app stays in control.
Building the Future of Interactive AI Systems
OpenAI's Agents SDK for building and orchestrating autonomous AI systems.
MCP and Cursor
MCP connects Cursor to external tools and data sources.
Chain of Thought Reranking
Rerank LLM answers by extracting and refining their reasoning chains.
Building langchain-salesforce
How I built a LangChain integration for Salesforce CRM.
Entropy-Based Chain of Thought Injection
Inject chain-of-thought when entropy shows the model is uncertain.
NumPy MCP Server
An MCP server that lets LLMs run NumPy numerical computations.
Building Halcyon-1B: A Creative Writing Model
Fine-tuning Gemma into a creative-writing model for storytelling.