Alison Hill

95% of AI engineering is just Context engineering. Everyone's obsessed with better models while context remains the real bottleneck. Even the best model in the world will give you garbage if you hand it the wrong information. Here's what most people miss: Context engineering isn't just about RAG or memory or agents. It's the art and science of delivering the right information, in the right format, at the right time, to your LLM. Think about what you actually need: ↳ Retrieval to fetch relevant documents ↳ Short-term memory to track conversations ↳ Long-term memory to remember user preference ↳ Agents to orchestrate everything ↳ Tools to extend capabilities That's 5 different systems you have to build, connect, and maintain. I've been building with Pixeltable recently, and it's an interesting approach to this problem. It's open-source and treats context engineering as a unified data problem: The idea is simple: instead of stitching together a vector database, a SQL database, an embedding service, and an agent framework, everything lives in one system. Your documents, embeddings, conversation history, and agent outputs are all just tables. Embeddings are computed columns that update automatically. Vector search works alongside your regular data operations. What I find useful: ↳ RAG pipelines without managing separate databases ↳ Long-term memory through vector search over historical conversations ↳ Multi-agent workflows that persist automatically ↳ Token budget management built into the framework It's not magical, but it removes a lot of the integration overhead. You're not fighting with three different APIs to make retrieval, memory, and agents work together. I've shared a starter notebook in the next tweet on building a context engineering pipeline with Pixeltable. It covers all the components and things we've discussed here. Everything is 100% open-source.