Featured
Instruments for working with LLMs day to day
Data-driven experience report on token and cost optimization for Claude Code. Settings, cache, model dispatch, and costly anti-patterns.
The agent memory market is taking off in 2026. A landscape review of tools, funding rounds, and technical trends.
The Agent2Agent (A2A) protocol standardizes direct communication between independent AI agents. Complementary to MCP, it covers the agent-to-agent relationship via HTTP/JSON-RPC 2.0, Agent Cards, and stateful task lifecycles.
Tool calling and MCP both connect an LLM to external tools, but at two different levels of abstraction. The first is a stateless API primitive, the second a stateful network protocol with dynamic discovery and exclusive primitives.
Claude Code is a development agent that runs in the terminal. A look at its architecture, features, and positioning against alternatives.
Two strategies for specializing an LLM, two opposing philosophies. One engraves knowledge into the weights, the other consults it on demand. The real debate isn't which is better, but which fits your use case.
MCP standardizes how language models access external tools and data. An open protocol, a client-server architecture, a fast-growing ecosystem — and unresolved security issues.
Prompting is the art of phrasing instructions for an LLM. A few simple principles make the difference between a mediocre answer and a useful one.
Retrieval-Augmented Generation (RAG) lets an LLM access data it never saw during training. Here is how it works.
A Claude Code skill is a self-contained instruction file that the AI reads before executing a task. Analysis of 116 community skills to extract the canonical structure and effective patterns.
Monitoring an LLM agent is not just infrastructure monitoring. An overview of specialized tools, the emerging OpenTelemetry standard, and areas that remain unresolved.