ALTK‑Evolve: On‑the‑Job Learning for AI Agents

ALTK-Evolve, published by IBM Research, is a memory system for AI agents that enables on-the-job learning by helping agents improve over time, learning from and using guidelines generated from previous executions. Operationally, it runs as a continuous loop: a capture layer records full agent trajectories, pluggable extractors mine traces for structural patterns and persist them as candidate entities, and a background consolidation job merges duplicates, prunes weak rules, and boosts proven strategies — evolving a high-quality library of guidelines, policies, and SOPs. By turning raw agent trajectories into reusable guidelines, the approach boosted reliability — especially on hard, multi-step tasks — achieving a 14.2% improvement on the AppWorld benchmark without bloating context.

Related

• Build knowledge agents without embeddings
• Chat SDK brings agents to your users
• Making Turborepo 96% faster with agents, sandboxes, and humans
• Enable this feature in the Cursor dashboard: http://cursor.com/dashboard/cloud-agents#my-pull-requests

Source: tools