RAG was never the end goal.Memory in AI agents is where everything is heading. Let me break down this evolution in the simplest way possible.RAG (2020-2023):- Retrieve info once, generate response- No decision-making, just fetch and answer- Problem: Often retrieves irrelevant contextAgentic RAG:- Agent decides if retrieval is needed- Agent picks which source to query- Agent validates if results are useful- Problem: Still read-only, can't learn from interactionsAI Memory:- Read AND write to external knowledg ...

Continual Learning in AI Agents - The industry recognizes the gap between AI agents and human learning capabilities, emphasizing the need for agents to continually learn and improve over time [1] - The industry is exploring different methods for AI systems to learn, including weight updates and learning in context using large language models (LLMs) [2] - Reflection over trajectories is emerging as a key theme, allowing agents to update memories, core instructions, and learn new skills [3][4][5] Skill Learning and Implementation - Skill learning involves reflecting over trajectories to learn skills, exemplified by the skill creator skill adapted from Anthropic [8][9] - Deep agent CLI allows specifying environment variables for logging traces, which is useful for reflection [10][11] - The industry is using Langsmith Fetch to grab recent threads from deep agents for reflection and persistent skill creation [12][13] - A practical example demonstrates how an agent can read a JSON file, reflect on its contents, and create a new deep agent skill, showcasing the utility of continual learning [15][16][17] Benefits and Future Directions - Skill learning enables agents to encapsulate standard operating procedures, such as grabbing Langsmith traces, for repeated use [19][20] - Continual learning loop involves agents reflecting on past trajectories to learn facts, memories, skills, and improve instructions [21][22]