New research shows AI agents don't just learn human capabilities — they systematically inherit human cognitive biases. The implications for deploying agents as objective decision-makers are uncomfortable.
Three independent papers demonstrate agents rewriting their own training code, generating their own knowledge structures, and refining their reasoning at test time. Self-improvement has moved from theory to working engineering.
Automated adversarial tools are emerging where small, cheap models systematically find vulnerabilities in frontier models. The safety landscape is shifting from pre-deployment testing to continuous monitoring.
Lab benchmarks show multi-agent systems coordinating well. Deploy them in messy reality and three kinds of friction emerge that no architecture diagram accounted for.
The next generation of agents will not be defined by peak capability but by their ability to match effort to difficulty. Across every subsystem, the field is converging on the same fix: budget-aware routing.
Twelve papers appeared on arxiv this week that, taken individually, look like incremental progress. Taken together, they describe a system architecture that did not exist six months ago: AI agents that rewire their own communication networks, embed their own auditors, and negotiate with each other for economic outcomes. The fixed-topology,
