"This year is the year zero of memory." That was the happiest sentence I heard yesterday. I am very honored to be invited to the roundtable of the Extraordinary Awards to have an in-depth discussion with several experts on the memory, architecture, and future of Agents. When it comes to building an Agent, memory is unavoidable. If used well, memory can make the Agent more proactive, make the Agent easier to use, allow the Agent to self-evolve, and give the Agent more of a "mother-born feel."
@elliotchen100
-
Claude Fable 5 Hints Future AI Competition Key Is Capability Boundary Design
By
–
Claude Fable 5 这次最值得看，不只是“更强模型发布”。
— 艾略特 (@elliotchen100) 10 juin 2026
而是 Anthropic 开始把最前沿能力拆成两层：大众可用的 Fable，加安全分类器；受信任场景里的 Mythos，开放更少限制。
AI 产品接下来竞争的关键，可能会从“谁的模型最强”，变成“谁能把能力、风险和分发边界设计清楚”。 https://t.co/t8vlcSB4KVClaude Fable 5 is the most noteworthy this time, not just because of a 'stronger model release'.
Rather, it's that Anthropic has started splitting cutting-edge capabilities into two layers: Fable, available to the public with a safety classifier; and Mythos, used in trusted scenarios with fewer restrictions.
The key to future AI product competition may shift from 'whose model is strongest' to 'who can clearly design the boundaries of capabilities, risks, and distribution'. -
From Prompt Optimization to Agent Loop with Memory
By
–
Peter put it well. The next stage is not "writing better prompts," but designing a loop that can continuously drive an agent. Let the agent observe, remember, retrieve, act, and improve with each run. But the loop needs memory. Without persistent, retrievable, and auditable memory, the so-called agent loop will eventually fall back to "cramming more context into the prompt."
-
AI memory requires engineering structure and data boundaries
By
–
AI memory is not something that can be explained with just a demo. It must eventually land in engineering structure, data boundaries, and real workflows.
-
EverOS and Skill Automation Through Use
By
–
5. Skill Automatic Accumulation EverOS extracts common skills from real usage patterns and consolidates repetitive behaviors into reusable workflows. I find this the most interesting layer of memory: it’s not about making the agent "remember more," but about teaching it how to perform tasks by learning from past experiences.
-
Elliot Chen on Fei-Fei Li’s “disciplined” approach to ‘world models’
By
–
After reading Professor Fei-Fei Li's piece, I felt like what she's doing is actually something quite restrained—taking a term that's been beaten to death and giving it some new discipline. Nowadays, everyone claims they're building a "world model," but video models, physics
-
Claude Code: Agent control planes, not just workflow generation
By
–
On the surface, it's Claude Code generating workflows on its own, but digging deeper, it's the control plane of agent products that's changing. In the past, we crammed complex tasks into a long context, expecting the model to remember the goal, break down steps, and judge
-
EverOS: redesigned interface with long memory for AI agents
By
–
我们新设计了 EverOS 的界面， 现在主要想突出两个点。
— 艾略特 (@elliotchen100) 30 mai 2026
第一，它是面向 AI Agent 的“长程记忆操作系统”。
传统 LLM 应用更多依赖短期上下文，而 EverOS 试图把 agent… pic.twitter.com/mwLmJLrFXjWe have redesigned the EverOS interface, and we now wish to highlight two main points. First, it is a "long memory operating system" designed for AI agents. Traditional LLM applications rely more on context to
-
AI Agent Memory Management and Data Structure Integration
By
–
好问题，这两个点很关键，安装本身不会改写原有 agent 的本地记忆文件；如果发现可导入的历史记忆，会先扫描并征得确认，再进入 EverMe 的结构化抽取/归类流程。 验证上，可以用一条已知历史事实跨 agent/session recall，并在 Memory Hub 里反查 source。如果未来停用，影响的是 EverMe
-
AI Agent Memory Persistence Across Coding Tools
By
–
To add: The "memory" mentioned here isn't just "chat history." Whether the agent can remember your projects, toolchains, preferences, pitfalls, and reusable practices long-term, and continue using them across Codex / Claude Code / Hermes. If you're working on agent workflows,