Blog.

Building a remote MCP server on Cloudflare Workers

How to scope, design, and deploy a Model Context Protocol server that connects Claude to internal tools through a secure, authenticated endpoint.

MCP

MCP agents and function calling: when to use each

How tool-calling agents differ from MCP-connected workflows, and where each pattern fits in internal AI systems.

MCP

How to run a workflow audit before you automate

Map inputs, decisions, tools, memory, outputs, review steps, and failure modes before choosing what to hand to an AI system.

Strategy

The three phases of AI deployment

A practical model for moving from workflow audit to pilot build to controlled rollout without forcing every engagement into the same timeline.

Strategy

RAG evaluation for internal knowledge bases

How to measure retrieval quality, answer grounding, and failure modes before trusting a RAG system in production.

RAG

Context engineering for large codebases

How coding agents get better when repo instructions, retrieval, dependency maps, and definitions of done are treated as infrastructure.

Agents

Using AI for code review with judge agents

How to build a judge-model loop that reviews pull requests for correctness, style, and risk without replacing the human in the loop.

Agents

AI governance for small and mid-size businesses

A lightweight governance framework for teams that need policy, access controls, and audit trails without an enterprise compliance team.

Strategy

LLM retrieval and grounded answers

How retrieval-augmented generation works in practice: chunking, embedding, retrieval ranking, and keeping answers tied to source documents.

RAG

Claude Cowork: what it is and how a team uses it

Claude in a browser answers questions about uploaded files. Cowork works in an approved local folder so the deliverable stays on disk.

Agents

Can Claude connect to CoStar?

There is no official CoStar connector. CoStar's terms bar sharing logins with AI agents, scraping, and exposing licensed data to generative AI. What to automate instead.

Strategy

Claude Skills vs MCP: when to use each

Skills package how Claude should do something; MCP grants access to do it. A practical guide to choosing between the two and when you need both.

Skills

How to write a SKILL.md that actually gets used

Why most Skills fail to trigger or bloat the context window, and the disciplines that make a Skill reliable: a sharp description, a lean body, and on-demand reference files.

Skills