Agents

// Programmatic Memory Enables Long-Horizon Reasoning // Keep the entire interaction log and search it. It works great and beats bespoke mem…

// Programmatic Memory Enables Long-Horizon Reasoning // Keep the entire interaction log and search it. It works great and beats bespoke memory harnesses on long-horizon tasks. New research introduces

DGX agentx-post
agentsdair-ai--x

// Programmatic Memory Enables Long-Horizon Reasoning // Keep the entire interaction log and search it. It works great and beats bespoke memory harnesses on long-horizon tasks. New research introduces PRO-LONG, a minimal context-management framework that gives LLM agents programmatic memory. It stores a complete, structured record of everything the agent has seen and uses coding-agent tooling to query that history on demand. The usual tradeoff is that saving more information makes the right detail harder to retrieve. PRO-LONG sidesteps it by treating history as a searchable database instead of something to compress. On the full ARC-AGI-3 public game set it improves over a base coding agent by 18.0 points on average across frontier models, and matches or exceeds specialized harnesses at up to 76.1% pass@1 while using 4.2 to 5.8x fewer tokens. With Fable 5 it reaches 97.4% best@2 for a total of $1,750. Paper: https://arxiv.org/abs/2607.20064 Learn to build effective AI agents in our academy: https://academy.dair.ai/

Related

Source: DAIR.AI (X) | 2026-07-23

Loading related sources…