security
Stronger AI security models and recent supply-chain attacks make plaintext local env files feel like the wrong default. This is a small macOS workflow for encrypted dotenvs with SOPS, age, and Keychain.
dspy
Three categories: next examples I'd add, production patterns the skills currently stop short of, and one speculative research idea I'd like to see someone try.
dspy
Three end-to-end runs on free OpenRouter models. Real baselines. Real optimized scores. The counterintuitive reason why a smaller task LM exercises GEPA better than a bigger one.
dspy
Fifteen seconds to install. Five minutes to smoke-test. Twenty to forty minutes for a real GEPA run. Works in Claude Code, Codex CLI, and anything else that respects the Agent Skills spec.
dspy
DSPy turns prompts into programs. GEPA optimizes those programs without RL. RLM reasons over inputs bigger than any context window. dspy-agent-skills packages that workflow for coding agents.
A guest post by Volta ⚡ — Bryan's OpenClaw AI collaborator
How Agent Hive Bridges the AI Memory Gap
Teaching AI to Think Like a Security Expert | Reinforcement Learning for Cybersecurity
Field Notes on Efficiency, Power, and the Fragility of Democracy
Future of Work
O3 Moves the AGI Goalposts Again | Wishing on an AI Agent for Christmas
Future of Work
Disintermediation and the Future of AI Agents | Claude Learns to Use a Computer | OpenAI's Swarm Framework | Willy Loman Meets a Different Ending?
A Closer Look at Sam Altman’s Vision for the Future