Observations in San Francisco
Everyone’s friendly and welcoming!
ai
There’s No Coming Back from That Waymo Ride
The wife and I experienced autonomous driving for the first time, and I can’t help but imagine utopias. How far until we see the signs of Banks’ Culture?
Thoughts on AI (March 8, 2026)
The societal change brought about by AI will be unprecedented. It’s not like the Industrial Revolution, or the invention of the camera, or personal computers, or the internet, although we desperately want it to be. Most of humanity isn’t ready for what’s coming.
Code Mode with Skills
An alternate proposal for Code Mode that replaces MCP servers and hosted execution with SKILL.md files and OpenAPI specifications.
Clawdbot Got Agent UX Right
Clawdbot gives LLMs arms and legs and unleashes their full potential.
Live and Let Live
LLMs haven’t killed programmers, and they aren’t useless either. Arguments at these extremes mostly miss the point. What matters is whether they can build things that work.
Notes on Recursive Language Models
RLMs are a novel way to keep the context of the main agent loop small, eliminating context rot for long-context tasks while reducing costs.
Notes on ForecastBench
ForecastBench continuously evaluates the performance of LLMs against an automatically generated, continuously updated set of forecasting questions.
Lessons on Context Engineering
Practical guidelines on context engineering, like having an append-only context, using response prefill to remove/force tools, setting up restorable compression strategies, and more.
Sandbox MCP: Enable LLMs to Run ANY Code Safely
A Model Context Protocol (MCP) server that lets LLMs run code safely in isolated Docker containers.
Provide Tools to Your LLM Agents with Model Context Protocol
Notes on using OpenAI Agents SDK’s MCP support to integrate DiceDB MCP.
Building a Model Context Protocol (MCP) Server in Go
My experience in building an MCP server for DiceDB using the MCP Go SDK.
I Tried Vibe Coding, and I Don’t Think I’m Going to Tab ⇥ My Way to a Million-Dollar Startup
The author is clearly trying to channel his existential crisis and crippled ambitions into a blog post.
An Even Playing Field
We need to democratize AI to save our democracies.
Eight Things the Indian Government Can Do about AI
What can the Indian Government do about artificial intelligence?
AI Traffic Cameras Were Always a Bad Idea
An amateur policy analyst attempts to explain why (disregarding controversies) installing AI traffic cameras in Kerala was a bad idea.
AI and APIs
What do the recent advancements in generative AI mean for APIs?
AI-Generated Spam Pull Requests
Someone tried to open pull requests to open source projects with AI-generated code.
ChatGPT Explained by ChatGPT
A conversation with ChatGPT about ChatGPT. Who are you?