The post highlights constraints, mechanisms, and factors influencing Agentic Engineering, emphasizing the types of bottlenecks we’re hitting and how GPU shortages are driving product changes.

We observe Agentic Engineering to amplify impact of power users increasing the gap to everyone else. This post explores approaches to teaching Agentic Engineering.

The autoresearch technique promoted by Andrej Karpathy is easily accessible in the pi coding agent.

Cupcake enables policy enforcement for AI coding agents like Claude Code, Cursor and OpenCode

This post explores Hunter, Skeptic, Arbiter prompting for finding vulnerabilities in code as an example of debate-style prompting.

This post explores MCP sampling and illustrates how important guardrails are to protect clients against malicious servers.

The codebases of MCP servers tend to be early-stage API wrappers, far away in quality from existing SDKs. Non-determinism of LLMs adds…

The posts explores three approaches to manual OpenTelemetry instrumentation for OpenAI calls in Langchain and LlamaIndex.

Using automatic instrumentation to quickly assess which APIs are called by popular Python LLM demo apps.

Emergency procedures aim at stabilizing the system in a degraded state. When used properly, they result in faster incident response and …