Field notes · AI & software
Useful answers before you build.
Practical decision guides drawn from building AI agents, knowledge systems, ecommerce assistants, and full-stack products end to end.
Engineering resources
How to Secure an MCP Server in Production
A 2026 MCP security checklist covering OAuth boundaries, token audience validation, SSRF, confused-deputy attacks, tool permissions, and untrusted output.
How to Evaluate AI Agents in 2026: Outcomes, Traces, and Tool Use
A production evaluation method for AI agents that tests environment outcomes, tool trajectories, permissions, recovery, and cost—not just the final answer.
Context Engineering for Long-Running AI Agents
A 2025–26 guide to managing instructions, tools, retrieval, memory, compaction, and subagents when an agent must work across many steps.
AI Agent Protocols in 2026: MCP, A2A, A2UI, AG-UI, and UCP
A practical guide to the 2026 agent protocol stack: which standards connect tools, agents, interfaces, and commerce—and when a normal API is still enough.
Agentic RAG in 2026: When Iterative Retrieval Beats One-Shot Search
A practical architecture for company knowledge agents that plan searches, route across sources, verify evidence sufficiency, preserve permissions, and stop on budget.
Stop Flattening PDFs: Multimodal Document AI in 2026
A production guide to preserving layout, tables, charts, images, and provenance when building AI systems over PDFs and mixed business documents.