AI Agents >= Interns The four agent design patterns Andrew Ng defined explained as if agents were just obsessive interns: 1. REFLECTION: The intern who returns with "I reviewed my first draft and found these issues, here’s the latest". They catch their own mistakes before you

Solid breakdown of memory - here's how I think about it Semantic memory: Facts or preferences (birthdays, names) usually user based Episodic memory: experiences with an understanding of previous context - previous conversations Procedural memory: Skills & knowledge execution -

If you're reading this, you're early

👀

Agent swarms require shared memory. We've been heads down building the past few months, and here’s what we’ve accomplished on project untitled Milestones: - Network running locally with payments & block production (via Ouroboros) - Reads & writes proofs - Read requests operational

Context windows are growing slower than agent swarm need

I speak English, APIs speak JSON LLMs speak XML?? I guess????

𝗔𝗜 𝗔𝗴𝗲𝗻𝘁’𝘀 𝗠𝗲𝗺𝗼𝗿𝘆 is the most important piece of 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴, this is how we define it 👇 In general, the memory for an agent is something that we provide via context in the prompt passed to LLM that helps the agent to better plan and

First, there were agents with context scoped to conversations. Useful agents. Now, there are agents with local memory, trapped to a single machine or a single company. Silo’d agents. Next, there will be agent swarms coordinating on shared memory. Collaborative agent swarms.

You said it perfectly. The future of AI is collective, not silo’d. That's why we're building the shared memory layer for agents. We're the team to do it. Welcome Christine Yip. LFG

You still build for humans? That’s cute

Played with veo3 with my mom this weekend “and this is the worst it’s going to be? It only gets better & faster from here?” …Yes…

Where do your agent’s context live? Locked into GPT? Gemini? Claude?  We’ve seen this movie before, you used Facebook for 15 years and all of your user data is owned by them, monetized by them, controlled by them. We’re not tolerating this again. We don’t have to.