Do Agents Actually Learn? Persistent Memory Experiments Show Mixed Results
artemgetman · reddit · 2026-10-24
artemgetman asks a fundamental agent-design question: can an agent improve at working with you without you constantly updating prompts, skills, and memory?
In his Cortex project during Anthropic's Claude Code hackathon, he experimented with persistent memory: some comparisons improved, others favored the baseline—so he's not convinced adding more memory solves it. He suspects adapting model weights could help with repeatedly used skills, but wants a rigorous comparison against a strong memory + skills setup.
His test is simple: give an experienced agent and a fresh copy unfamiliar work—does the experienced one get better results, need fewer corrections, and retain what it learned? The key question is whether learning transfers across genuinely different tasks, and whether the agent recognizes when a comparison doesn't apply.
Related event: Developers Question Whether AI Agents Truly Learn(2 posts)→
More from coding & agent
- YCloud ships WhatsApp API MCP server letting Claude chat with WhatsApp directly — modelcontextprotocol · 2026-10-24
- NoonAI DIS MCP server auto-detects and redacts faces in images and video — modelcontextprotocol · 2026-10-24
- Google Research tests 180 agent setups: multi-agent hurts sequential tasks by 39-70% but boosts parallel ones 80.9% — ombre-ai · 2026-10-24
- OpenAI Skills appear only in ChatGPT Work, not Chat, user reports — mark_k · 2026-10-24
- 6 bugs that break naive LLM budget caps under concurrency—and their fixes — MaverikSh · 2026-10-24
- Use Your Own Org Data to Predict Slow Cases Before Customers Complain — louisdorard · 2026-10-24