Agent Frameworks Considered Harmful — Rémi Louf, .txt
Summary
The talk details the creation of a robust, event-driven agent runtime system designed to overcome limitations in existing AI frameworks and chat interfaces. The core innovation is moving from string-based prompts and complex graphs to an auditable, content-addressed store where every component (system message, skill description, user question) is hashed and stored separately. This architecture ensures full traceability, allowing for precise diffing between runs and reliable replaying of requests against different models, making the system highly debuggable and scalable.
Key takeaways
-
Event-Driven Architecture (EDA)
15:37
The system relies on agents subscribing to typed events rather than maintaining complex graphs with edges. This simplifies orchestration significantly, allowing for 'funneling' and 'fan out' without requiring code knowledge; users only need to know what events exist in the system.
-
Content-Addressed Prompt Store
20:11
Prompts are no longer rendered strings. Instead, every part (system message, skill description, tool definition, user question) is hashed and stored as an identifier. A prompt is represented as a list of these hashes, enabling precise auditing.
-
Enhanced Observability
The system implements two key boundaries: typed tool calls and typed events. This rigorous typing prevents malformed or non-existent inputs from breaking the pipeline, which was necessary because early failures included rejected/malformed events.
Technical details
-
Agent Orchestration
611s
The system uses a declarative approach where agents are defined as markdown files, allowing non-coders to contribute. The runtime handles the execution flow based on event emissions (e.g., a voice note agent emits 'voice_note_processed', which triggers a daily brief agent).
-
Prompt Versioning and Diffing
1211s
By treating prompts as graphs of hashes rather than rendered strings, the system allows users to perform accurate diffs between runs (e.g., identifying if only the user message changed) and reliably replay old requests using different models.
-
System Logging
937s
A central, append-only events table serves as the system's memory. All events are causally linked, which is crucial for debugging complex multi-agent workflows and preventing data loss (e.g., lost notes or uncounted attempts).
Mentioned resources
- The Typical Set
- Blog Post on Typed Events
Channel & topics
Watch on YouTube · Back to latest
This independent, AI-assisted summary is provided for commentary and informational purposes. It may contain errors or omit important context. Please watch the original video for the creator's complete presentation. Video, thumbnail, and related copyrights belong to their respective owners.