The Agent Architect #31: Sliding Window Management

Pattern of the week

Sliding Window Management

What:

Dynamically shrinks or expands the context window during execution, prioritizing recent and relevant information while dropping less important older content.

When to use it:

Long-running tasks where full conversation history exceeds token limits and you need to preserve task-critical information while dropping tangential details.

Watch out:

Relevance scoring is heuristic-based; you may accidentally drop context needed later, causing the agent to repeat work or lose important constraints.

This week in agentic AI

• A corrective agentic hybrid RAG and an operations-grounded evaluation for a scientific facilityarXiv cs.AI

  APS-RAG is a deployed retrieval-augmented generation platform that integrates decades of operational knowledge from scientific facilities including logbooks, documents, and control-system data.

• Agentic Permissions Policy Algebra for Taint Confinement in LLM AgentsarXiv cs.AI

  APPA is a permissions policy algebra for LLM agents that enables selective taint tracking to balance security against prompt injection while maintaining utility.

• Opaque Epistemic Mediation: How LLM Deployment Configurations Shape the Validation of Pseudo-SciencearXiv cs.AI

  Four major LLM families show inconsistent and non-transparent stances on contested scientific claims across multiple time snapshots.

• CausalForge: A Formally Grounded, Self-Improving Agentic Framework for Automated Research in Causal InferencearXiv cs.AI

  Framework automates causal inference research by using formal verification instead of LLM reviewers to evaluate candidate results.

• TRACE-ROUTER: Task-Consistent and Adaptive Online Routing for Agentic AIarXiv cs.AI

  Router selects LLMs for agentic workflows by considering task-level outcomes rather than making independent per-call decisions.