Working memory

As explained in The Art of AI Product Development

Agent memory held within the model's current prompt context and available for immediate processing.

Working memory is the information an agent keeps in its current prompt context for immediate processing. It preserves continuity across multistep work, such as carrying trend research into roadmap suggestions, even though language models otherwise reset with each prompt. section 9.2.3

Working memory may contain:

• Interaction history, including conversations and human-in-the-loop inputs. section 9.2.3
• Personalization details, completed tasks and outcomes, or retrieved data from RAG sources. section 9.2.3
• An activity traceback, synthesized summary, or selected relevant thoughts and interactions. section 9.2.3

Roadmap research

An agent researching trends can place its findings in working memory so a later step can synthesize roadmap suggestions. Designers must balance retained usefulness against context size: larger prompts increase latency, token count, and potential inaccuracies. section 9.2.3

Figure 9.14 Agents need short-term memory for the current session and long-term memory for learning and improvement over time.

Compare working memory for the current session with long-term memory stored externally for later retrieval and improvement. section 9.2.3

How The Art of AI Product Development covers it

How deep it goes

Explains stored content, prompt representations, and context trade-offs, but stops short of implementation details.

Examples

Uses chatbot continuity, trend research, roadmap synthesis, conversations, task outcomes, and retrieved data as examples.

What it assumes

Assumes multistep agent workflows and teaches why stateless language models need prompt-based continuity.

What you can do after

Operate agent workflows by selecting useful context while managing latency, token count, and potential inaccuracies.

How it connects

Connects working memory to prompt context, context windows, long-term memory, RAG databases, and agent systems.

liveBookmanning.com

Read The Art of AI Product Development on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  AI Agents in Action, Second EditionShort-term memory is the recent interaction history of the current session that remains directly in the model's context window.
• 

  liveBookmanning.com
  Building Reliable AI SystemsAn agent memory that temporarily retains information, typically for the duration of a single session.
• 

  liveBookmanning.com
  Natural Language Processing in Action, Second EditionShort-term memory is a temporary store for information that remains available over a limited span of processing.
• 

  liveBookmanning.com
  Domain-Specific Small Language ModelsConversation memory that keeps recent interactions within a session so they can be reused immediately.
• 

  liveBookmanning.com
  Inside AIWorking memory is the component of an expert system that stores input data and intermediate facts produced during reasoning.
• 

  liveBookmanning.com
  Introduction to Generative AI, Second EditionShort-term memory is an agent capability for retaining relevant intermediate information from the current task or interaction.

Compare how 7 books explain working memory →

Related concepts

• memorykind of · Memory is a storage resource or mechanism that retains data or computational state for access during computation or later use.
• context windowequivalent to · A context window is the bounded sequence of tokens or other input information a model can access when processing or generating an output.
• statecontrasts with · State is information or a configuration describing the condition of a system, entity, or computation at a particular time and potentially influencing subsequent behavior or decisions.
• cognitive loadused by · Cognitive load is the mental effort required to process information, retain it in working memory, or perform a task.
• cognitionkind of · Cognition is the set of processes by which an agent acquires, processes, represents, and uses information.
• trajectoryuses · A trajectory is an ordered sequence of states, actions, events, or operations that records or describes progression through successive steps or over time.

This explainer was generated by Manning's knowledge graph from the text of The Art of AI Product Development. The section links point to the parts of the book it draws on.