Yohei Nakajima, creator of BabyAGI, introduces ActiveGraph, a novel agent architecture that builds around an immutable event log rather than the LLM. This log-centric approach enables auditable agents with native replay, rollback, and self-improvement capabilities, drawing inspiration from blackboard systems and Kafka.
Dan Farrelly, CTO of Inngest, addresses the rapid obsolescence of AI agent architectures (a 6-month half-life) caused by fast-evolving models and frameworks. He proposes a solution: decouple agent systems into three conceptual layers (Execution, Context, Compute) and prioritize a stable, durable Execution Layer. This 'brain' layer, responsible for flow, state, and retries, offers resumability, flexible invocation patterns, and comprehensive observability, allowing the 'knowledge' and 'hands' layers to change frequently without necessitating full architectural rewrites.
Shahram Anver, CEO of Cleric, details how AI agents for SRE are evolving beyond fast triage to continuous learning and operational memory. He discusses Cleric's architectural shifts, from complex early designs to simpler, sandboxed query agents, and the unique challenges SRE agents face in diverse production environments. A core focus is on human-agent interaction, redefining roles as managers overseeing agents, and how agents learn from unstructured data like Slack to build robust, actionable knowledge for autonomous, self-healing infrastructure. The discussion also touches on the future of software, differentiating between durable systems and rapidly developed "skills" or "vibe-coded" solutions.
Gal Peretz introduces CodeAct, a paradigm where LLMs generate and execute Python code for tool interaction, offering a more flexible and powerful alternative to traditional JSON-based function calling for building complex, production-ready AI agents.