The latest trail, Agent Memory
Season one ran four essays, from whether we even mean the same thing by “memory” to how agent memory forms, updates and fades. Season two asks whether an agent that remembers also remembers right.
Everyone Talks About Memory. Do We Mean the Same Thing?
Everyone working on AI agents talks about memory, but chat history, MEMORY.md, RAG, context engineering and fine-tuning answer different questions. The first essay in my Agent Memory series pulls them apart.
1,881 words2026.08.24 S1·2What Does an Agent's Memory Actually Look Like?
Where an agent's memory lives and what form it takes are two different questions. A tour of token-level, parametric and latent memory, and where MEMORY.md, vector DBs and knowledge graphs fit.
2,601 words2026.08.24 S1·3What Should an Agent Remember?
Not everything that happens is worth an agent remembering. Factual, experiential and working memory each do a different job, and stacking them against memory's form gives you a map of the field.
3,027 words2026.08.24 S1·4How Agent Memory Forms, Updates and Fades
Memory isn't a static snapshot. It has a lifecycle of formation, evolution and retrieval, with consolidation, updating and forgetting in between. The season one finale of my Agent Memory series.
3,230 words2026.09.04Remembering Isn't the Same as Remembering Right
Once an agent stops forgetting, the next problem is whether it keeps remembering wrong. How to build reliability into writing, updating and reading memory, and how to test that it actually worked.
3,936 words2026.09.22What if someone knows your agent trusts its memory, and plants something false on purpose?
In the works
92 essays, by region
Regions are the place names on the map above; pick one and its cairns light up.
Nothing in this region yet.
27 talks on this year’s trail
An ascent profile of talks given this year. Marked points are major events; the dashed line is still ahead.
WiDS Taiwan (Women in Data Science)
Five AIs on My Team: Building Your Agent Team
Recap →AWS Summit Taipei 2026
From AI Sessions to Shared Memory: Knowledge Infrastructure for AI-Native Teams
Recap →Agile Taiwan
From OpenClaw to Cairn: When Agents Join the Team, How Should Memory Be Shared?
Recap →Taiwan AI Annual Conference 2026
Corporate training
Details soon
Talk for a startup development association
Details soon
Things I’ve built
Two products, plus open-source projects that usually come with a tutorial.
Cairn
A cairn is a pile of stones hikers leave to mark the way for whoever comes next. This map borrows the idea behind Cairn: what each person leaves behind becomes a path someone else can find.
Shared memory for people and AI agents. People leave real experience as field notes, and Moss, a custodian agent, organizes, links and maintains them. Built solo in six days for Anthropic’s global hackathon.
Build log → AILogora
A Chinese-first community for AI practitioners, organized around viewpoints and topics: perspective cards, discussions, a knowledge map, and a weekly digest for nearly 4,000 subscribers.
Visit AILogora →LLMAvatarTalk
A real-time voice assistant with an animated avatar, built on NVIDIA Riva ASR/TTS, LangChain and Audio2Face. Top 100 in the NVIDIA × LangChain Generative AI Agents Developer Contest.
GitHub →GraphRAG Visualization Tutorial
A hands-on guide to Microsoft GraphRAG with visualizations and a prebuilt index, so you can explore it without paying for indexing.
GitHub →cairn-memory
The open-source memory layer behind Cairn: cross-session memory for AI agents where every memory carries a receipt, the exact source text it came from. Inspect it, correct it, or forget it for good.
GitHub →RAG_LangChain_streamlit
A from-scratch retrieval-augmented generation example with Streamlit, LangChain and Llama 2.
GitHub →LLM Spec Comparison
A side-by-side comparison of mainstream LLMs: ChatGPT, Gemini, Claude, Mistral, Llama and more.
GitHub →skill-openclaw-map
A map of OpenClaw skills and workflows, gathered into one place you can actually search.
GitHub →On OpenClaw and AI agents, in the press
Interviews about OpenClaw, an open-source AI agent, and what it takes to use agents for real.
2026 Business Weekly
Interviewed on OpenClaw and AI agents in practice: what they are good for, how hard they are to start with, and what I have seen.
2026 USTV News
A feature interview demonstrating AI agents for admin work (slides, meeting notes, an agent round table), plus the risks and what NemoClaw changes.
2026 PTS (Taiwan Public Television)
A news interview and a PTS talk-show appearance, sharing daily life with AI agents as an OpenClaw user.
▲ SURVEYOR I’m ChiChieh Huang, a founder and AI engineer in Taipei.
I’m building Cairn, shared memory for people and AI agents. Before that I built AILogora, a Chinese-first community for AI practitioners that went through AppWorks Accelerator #31. Since late 2023 I’ve written 92 essays explaining generative AI in Chinese, and in 2026 I gave 27 talks across Taiwan.
- Focus
- LLM applications, RAG, AI agents, agent memory, speech, 2D/3D avatars
- Work with me
- Talks, workshops, corporate training, advisory
- Contact
- [email protected]
- More
- Full bio GitHub LinkedIn RSS