Urgent.News

What's breaking now, across hundreds of outlets.

Tech

The Components of Agent Memory

Article URL: https://www.anthropic.com/research/components-of-agent-memory Comments URL: https://news.ycombinator.com/item?id=50402212 Points: 169 # Comments: 99

The concept of memory in AI agents is a complex one, drawing from the workings of human cognition. Just as humans encode and consolidate memories, AI agents such as Claude can capture experiences and store them for later use. However, not all of these experiences make it to long-term memory. During periods of rest, the human brain replays recent experiences, a process that stabilizes and integrates them into long-term knowledge storage.

Similarly, AI agents can use idle time to "dream" or revisit their recent experiences, helping to organize their memory. One way this is implemented is by tagging important events during a session and saving them to a memory store for later reference. Claude, for instance, can write down salient events as they happen or tag them in a log.

This "dreaming" process allows Claude to reflect on its past actions, correct mistakes, and predict which information will be relevant in the future. This predictive aspect of memory formation is crucial, as it enables agents to anticipate potential needs based on past patterns. For example, when given a task to answer questions based on a database, Claude can write down rules or notes that it believes will be useful in future instances of the same task.

This self-learning mechanism, akin to continuous learning in humans, allows Claude to improve its performance over time. Evaluations of Claude's memory capabilities have shown varying results. While some models like Claude Opus 5.5 and Claude Sonnet 5.5 have performed well by consolidating information and marking verified facts, others like Claude Sonnet 3.7 have written extensively but not effectively, resulting in fewer badges in challenging tasks such as playing Pokémon Red.

This highlights the importance of a well-organized memory system in AI agents, where only relevant and valuable information is retained. The challenge for developers lies in creating a system that knows what to remember and what can safely be discarded, much like how the human brain prioritizes what is important during offline periods of rest.

Rewritten by urgent.news; may contain errors — check the original.

This story

This is one outlet's version. Read the fullest account.

Every outlet that reported this, and when →

Read the original at anthropic.com →

More in Tech

Treat a crawler skill's output as a versioned interface

A saved crawler workflow is easier to reuse when the next person can trust its output shape. Finding the same button again is only part of the job.

  • Define contract before saving skill to prevent downstream issues
  • Specify key fields: source ID, URL, email handling, row limits
  • Use Python validator to check contract compliance

Ming-3.0-tts on a 16 GB card: the GPU sat mostly idle, and why a bigger one barely helps

When a speech model runs slower than real time, the reflex is to rent a bigger GPU. With Ming-3.0-tts-0.5B that buys very little. An RTX 2000 Ada sat idle more than two thirds of the time.

  • RTX 2000 Ada GPU idle >2/3 time while processing Ming-3.0-tts
  • RTX 4090 only effective once in five instances, 17% speedup
  • 16 GB card insufficient; 4 GB estimated minimum for stable performance

My Analytics Agent Stopped Repeating Its Wrong Guesses With Hindsight

My Analytics Agent Stopped Repeating Its Wrong Guesses With Hindsight Every Monday morning, someone in leadership asks why a regional revenue metric dropped, and an analyst spends two hours checking…

  • WhyDesk replaces LLM-driven tool with Hindsight methodology
  • System uses Vectorize agent memory to prioritize efficient diagnostics
  • Three memory types: world, experience, observation

More from Saturday 7 November →