Welcome! This is where I write about building AI systems that actually ship: AI agents, MCP servers, retrieval, and the AWS infrastructure underneath them.
Why another blog?
I started out writing on Medium, and while I love the platform’s reach and community, I wanted a place I fully control: somewhere I can experiment with format, own my content, and not worry about paywalls getting in the way. This site is the canonical home for everything I write. Medium still gets cross-posts, each one pointing back here.
What to expect
Three kinds of posts, all drawn from production work:
- AI agents and MCP: tool design, security, evaluation, and the failure modes that only show up after shipping
- Retrieval and RAG: benchmarks over folklore; more than one post here exists because a benchmark overturned what I expected to be true
- AWS and serverless: the infrastructure these systems run on, with real cost numbers attached
Much of it comes from two open-source MCP servers I build and maintain, pdf-mcp and redmine-mcp-server. When something breaks in production, or a benchmark settles a design argument, it usually becomes a post. If you want a starting point, the production AI agent playbook maps the agent posts, and the RAG decision guide maps the retrieval ones.
If you want to get in touch, find me on GitHub or LinkedIn.
Thanks for reading!
Kevin
