ESAA: Event Sourcing for Autonomous Agents in LLM-Based Software Engineering
#Event Sourcing #Autonomous Agents #LLM-Based Software Engineering #Deterministic Orchestration #Multi-Agent Systems #AI Verification #State Management #Forensic Traceability
📌 Key Takeaways
- ESAA architecture separates agent cognition from state mutation
- System uses structured JSON intentions and deterministic orchestration
- Two case studies validate the architecture including multi-agent scenarios
- Approach ensures immutability and forensic traceability through replay verification
- Architecture demonstrates scalability with heterogeneous LLMs
📖 Full Retelling
Elzo Brito dos Santos Filho introduced the ESAA (Event Sourcing for Autonomous Agents) architecture for LLM-based software engineering in a paper published on arXiv on February 26, 2026, addressing critical structural limitations in autonomous agents including lack of native state, context degradation over long horizons, and the gap between probabilistic generation and deterministic execution requirements. The ESAA architecture separates an agent's cognitive intention from a project's state mutation, drawing inspiration from the Event Sourcing pattern. In this system, agents emit only structured intentions in validated JSON format, while a deterministic orchestrator handles validation, persistence in an append-only log, application of file-writing effects, and projection of a verifiable materialized view. The architecture incorporates several innovative features including boundary contracts, metaprompting profiles, and replay verification with hashing to ensure immutability of completed tasks and forensic traceability. The paper presents two case studies that validate the ESAA architecture: a landing page project consisting of 9 tasks and 49 events with single-agent composition, and a more complex clinical dashboard system involving 50 tasks, 86 events, and 4 concurrent agents across 8 phases. Both case studies concluded with successful verification, with the multi-agent scenario demonstrating real concurrent orchestration with heterogeneous LLMs including Claude Sonnet 4.6, Codex GPT-5, Antigravity/Gemini 3 Pro, and Claude Opus 4.6, providing empirical evidence of the architecture's scalability beyond single-agent scenarios.
🏷️ Themes
Artificial Intelligence, Software Architecture, Event Sourcing
Entity Intersection Graph
No entity connections available yet for this article.
Original Source
--> Computer Science > Artificial Intelligence arXiv:2602.23193 [Submitted on 26 Feb 2026] Title: ESAA: Event Sourcing for Autonomous Agents in LLM-Based Software Engineering Authors: Elzo Brito dos Santos Filho View a PDF of the paper titled ESAA: Event Sourcing for Autonomous Agents in LLM-Based Software Engineering, by Elzo Brito dos Santos Filho View PDF HTML Abstract: Autonomous agents based on Large Language Models have evolved from reactive assistants to systems capable of planning, executing actions via tools, and iterating over environment observations. However, they remain vulnerable to structural limitations: lack of native state, context degradation over long horizons, and the gap between probabilistic generation and deterministic execution requirements. This paper presents the ESAA (Event Sourcing for Autonomous Agents) architecture, which separates the agent's cognitive intention from the project's state mutation, inspired by the Event Sourcing pattern. In ESAA, agents emit only structured intentions in validated JSON ( this http URL or this http URL ); a deterministic orchestrator validates, persists events in an append-only log ( this http URL ), applies file-writing effects, and projects a verifiable materialized view ( this http URL ). The proposal incorporates boundary contracts ( this http URL ), metaprompting profiles , and replay verification with hashing (esaa verify), ensuring the immutability of completed tasks and forensic traceability. Two case studies validate the architecture: a landing page project (9 tasks, 49 events, single-agent composition ii) a clinical dashboard system (50 tasks, 86 events, 4 concurrent agents across 8 phases), both concluding with this http URL =success and verify_status=ok. The multi-agent case study demonstrates real concurrent orchestration with heterogeneous LLMs (Claude Sonnet 4.6, Codex GPT-5, Antigravity/Gemini 3 Pro, and Claude Opus 4.6), providing empirical evidence of the architecture's scalability beyo...
Read full article at source