Prompts: the instructions everything else obeys
What prompt engineering still means in 2026, why the system/user split saves real money, and how to spot tools that hide the instruction layer from you.
Learn AI — one element at a time
The 20 elements of the RXed AI Periodic Table, explained in plain language for people who run a business, not a lab: what it is, when you actually need it, and which audited tools do it well.
Embeddings: how AI finds meaning in your files
Meaning as numbers: the trick behind semantic search and every chat-with-your-documents tool — and why it costs almost nothing.
Meaning as numbers: the trick behind semantic search and every chat-with-your-documents tool — and why it costs almost nothing.
Context: the working memory you have to manage
Why bigger context windows didn't kill context engineering, and how deciding what the model sees cuts cost and improves answers at once.
Why bigger context windows didn't kill context engineering, and how deciding what the model sees cuts cost and improves answers at once.
Tracing: the flight recorder your AI needs
Agents fail politely — well-formed, confident and wrong. Traces are how you find out why, and only ~15% of deployments have them.
Agents fail politely — well-formed, confident and wrong. Traces are how you find out why, and only ~15% of deployments have them.
The LLM: choosing the engine (it's a buyer's market)
Match the model class to the job, not the marketing — and fear the switching cost of everything around the model, not the model itself.
Match the model class to the job, not the marketing — and fear the switching cost of everything around the model, not the model itself.
Function calling: how AI stops talking and starts doing
The structured-action loop behind every agent that books, files or sends anything — and where MCP fits next to it.
The structured-action loop behind every agent that books, files or sends anything — and where MCP fits next to it.
Vector stores: where meaning gets indexed
Probably you just need pgvector: the boring 2026 consensus on storing embeddings, and the export trap to avoid.
Probably you just need pgvector: the boring 2026 consensus on storing embeddings, and the export trap to avoid.
RAG: making AI answer from your knowledge, not its guesses
The retrieval pattern behind every serious knowledge assistant — why chunking beats model choice, and the three questions that expose weak RAG.
The retrieval pattern behind every serious knowledge assistant — why chunking beats model choice, and the three questions that expose weak RAG.