The community is building solid infrastructure pieces today. Google Agent Skills signals official vendor support for agent tooling, while projects like Superlog tackle the observability gap that's been bugging teams running LLM workloads. The energy feels practical rather than flashy, people shipping tools we actually need rather than chasing demos.

Agent Skills are standardised capabilities that AI systems can invoke to perform specific tasks, like querying BigQuery or managing Kubernetes. They're appearing everywhere because they solve the integration problem between AI and real systems. Google's repo offers 20+ skills, LibreChat now supports them, and Prompt Refine creates them dynamically.

Stigmergic coordination is indirect agent communication through shared environment traces, like ants leaving pheromone trails. Instead of agents directly messaging each other, they leave evidence of their work that other agents discover and build upon. Axocoatl uses this for multi-agent systems where agents activate based on dependency completion.