Steven Mays
AI & Backend Engineer. 15+ years in production systems. Now shipping LLM agents, MCP servers, and agentic pipelines.
01 / How I Build
Multi-step agents as explicit state machines — not chained prompts. Each state declares its prompt, retry policy, and cost ceiling. Failures move to recovery states with known semantics. Resumable, observable, and easy to budget.
Saga orchestration, circuit breakers, idempotent handlers, event-driven pipelines that absorb cascading failures at thousands of events per second.
MCP servers and tool definitions that expose model-shaped views of internal data — ERP, product, analytics. Surfaced through connectors, Slack and purpose-built UIs. RAG and persistent agent memory where they actually pay off.
02 / By the Numbers
04 / Latest from the Blog
My stock-research agent evaluates one company at a time, fans out a dozen subagents to verify every claim, and runs only when I have spare tokens to burn. Slow by every metric that matters for an agent — which is exactly why I trust it. How it is built, and the first stock it picked.
Customize Claude Code with skills, plugins, commands, and configuration files that transform a capable coding assistant into one that matches your exact workflow.
How I fixed intermittent CORS failures and timeout issues in my AI tools by setting up a custom API subdomain and migrating from free to paid AI providers.
How I built a retrieval-augmented generation system that lets you chat with Warren Buffett's shareholder letters - to understand RAG trade-offs in production.