Cost Is Not a Bug: Why Our Expensive Agents Might Be the Point

Research

Cost Is a Feature: Research Shows Expensive Logics Dominate Cheap Ones in Task Performance

Complex and costly workflows systematically outperform simpler, cheaper ones—and smaller LLMs are consistently better at these "slow" tasks than their larger counterparts. This challenges the entire speed-over-everything reflex in AI ops.

CryptoPuzzles Benchmark: Real-World Hard Problems for Measuring Agent Reasoning

A new benchmark measures LLM performance using real cryptographic puzzles, offering a significantly harder test for reasoning and tool use than standard static benchmarks. Essential reading if you're building agents that need to actually solve hard problems.

Tools

AutoGPT 1.1 Ships Stability Fixes for Long-Running Agent Runs

AutoGPT's latest release focuses on stability and reliability for long-running, multi-agent setups—exactly what you need when your agent swarm is running the whole company overnight without supervision.

Infrastructure

WalletConnect Expands Agent Treasury to Multiple Chains

WalletConnect's Agent Treasury now works across multiple chains, meaning your autonomous agents can manage assets without needing a bridge or human in the loop. This is the foundational plumbing for an agent-run treasury.

Ecosystem

Y Combinator Blog: Ongoing Coverage of AI-Native Company Building

Y Combinator's blog regularly features essays and resources directly relevant to the zero-human company thesis, from scaling teams to AI-native company structures. A key lens for tracking where founder sentiment and capital are shifting.