Intermediate Artifacts as First-Class Citizens: A Data Model for Durable Intermediate Artifacts in Agentic Systems

2026-05-12

AI Summary

This work addresses the frequent loss of intermediate artifacts in multi-step agent systems, which hinders traceability and iterative refinement of outputs. It is the first to systematically distinguish such artifacts from dialogue logs, memory, or chain-of-thought reasoning, proposing a data model that treats them as first-class, persistently managed entities. The model introduces typed structures, explicit state resolution, and artifact lineage tracking, enabling structured, addressable, versioned, and dependency-aware maintenance, along with additive and overwrite update semantics. By shifting evaluation focus from final outputs to the quality of intermediate states, this approach establishes an artifact ecosystem that humans and agents can jointly inspect, revise, and reuse, substantially enhancing the verifiability, editability, and maintainability of AI-generated processes.

Abstract

Many AI systems are organized around loops in which models reason, call tools, observe results, and continue until a task is complete. These systems often produce final artifacts such as memos, plans, recommendations, and analyses, while the intermediate work that shaped those outputs remains ephemeral. For multi-step, revisable AI work, final artifacts are often lossy projections over upstream state. We argue that such systems should preserve durable, inspectable intermediate artifacts: typed, structured, addressable, versioned, dependency-aware, authoritative, and consumable by downstream computation. These artifacts are not the model's private chain-of-thought. They are maintained work products such as evidence maps, claim structures, criteria, assumptions, plans, transformation rules, synthesis procedures, unresolved tensions, and partial products that later humans and agents can inspect, revise, supersede, and improve. The contribution is a systems-level data model. We distinguish intermediate artifacts from chat transcripts, memory, hidden chain-of-thought, narration, thinking, and final answers; formalize additive and superseding update semantics with explicit current-state resolution; describe how artifact lineage supports durable intermediate state across revisions; and argue that evaluation must target maintained-state quality, not only final-output quality. The claim is not that artifacts make models smarter. It is that durable intermediate artifacts make AI-generated work more inspectable, revisable, and maintainable over time.

Problem

• intermediate artifacts
• agentic systems
• durable state
• inspectability
• maintainability

Innovation

• intermediate artifacts
• agentic systems
• data model
• artifact lineage
• durable state