Writing

Blog

Agentic engineering, spec-driven workflows and shipping real software. 22 posts.

  1. Shipping a Remote MCP Server That Works in Both ChatGPT and Claude: A Field GuideHow to ship a remote MCP server that ChatGPT and Claude both accept: tool design, JSON-RPC gotchas, OAuth 2.1, security and a two-host test matrix.
  2. Who Signs Off When Agents Write the Code? Risk-Tiered Approval in an AI-First SDLCAn AI-first SDLC approval workflow that grades each spec by risk, lets a reviewer agent sign low-risk work and sends sensitive changes to a human.
  3. Layered AI Code Review: Commit-Time, PR-Time and Human Review That Don't Step on Each OtherAn AI code review workflow in three layers: a local reviewer on every commit, a PR reviewer that owns the verdict, and humans for judgement calls.
  4. Bounded Autonomy: How to Let a Coding Agent Push Fixes Without Letting It ShipAI coding agent guardrails that let an agent push review fixes on its own: armed autonomy, a cycle budget, a deterministic guard and a clean hand-back.
  5. Separate Talking From Acting: A Guardrail Architecture for Production AI AgentsAI agent guardrails that hold in production: stage every turn, make tools the only write path and guard what the agent promises before it's sent.
  6. Your CLAUDE.md Is Probably a Changelog: Agent Instruction Files That ScaleCLAUDE.md best practices for long-lived repos: audit your agent instruction file, move history out, split decisions into ADRs, and generate status.
  7. LLMs at the Edges: Designing Conversational Agents With a Deterministic CoreHow to build a deterministic conversational agent with LLM extraction at the edges: a pure flow runner, cheap models for parsing and human takeover.
  8. ADRs and Session Handoffs: Giving AI Coding Agents a Memory That SurvivesArchitecture decision records for AI coding agents, plus a session handoff file, let a fresh session pick up fast and stop re-arguing settled choices.
  9. Evaluate What Your Agent Does, Not What It Says: A Guide to Tool-Call EvalsAI agent evals that check tool calls, not prose: expected tools, forbidden tools, grounded facts, and guardrails that prune claims no tool backed.
  10. RAG Over Your Own Docs With pgvector: Chunking, Retrieval and Answers You Can TrustA practical RAG pgvector guide for your own docs: chunking, hybrid search, citations, index freshness and evals that teach the bot to say I don't know.
  11. Human Handoff for AI Chat Agents: A Practical Design GuideAI agent human handoff, done right: what a chat agent may do alone, what it only drafts, and how to pass a conversation to a person without losing context.
  12. Red-Teaming Your LLM Agent: Prompt Injection and Prompt Leakage TestsA practical guide to prompt injection testing for LLM agents: leakage probes, indirect injection via RAG, image exfiltration and a CI gate.
  13. How to Build an Eval Gym for Conversational AI AgentsA practical guide to AI agent evaluation: four eval layers, LLM-as-judge scoring, N-run noise checks and simulated customers for chat agents.
  14. A Week of Spec-Driven Development With GSDAn honest week-long diary of shipping features on a real project with GSD. Phases, commits, time spent, and what broke along the way.
  15. Migrating From OpenSpec to GSD: When It's Worth ItThinking about switching from OpenSpec to GSD? Here's an honest look at what you gain, what you give up, and when the migration is worth the cost.
  16. What Is Spec-Driven Development? A 2026 PrimerSpec-driven development (SDD) is a methodology where specs come before code. Here's what it is, why AI agents made it practical, and how to start.
  17. GitHub Spec Kit: A Hands-On Review for 2026GitHub Spec Kit review after shipping real features with it. The Constitution, the 7 phases, the ceremony tax, and when it's worth the cost.
  18. OpenSpec Explained: A Hands-On Guide for 2026OpenSpec is a lightweight spec-driven development framework for brownfield codebases. How it works, when to use it, and where it falls short.
  19. Inside GSD 2.0: The Architecture Behind Reliable AI AgentsA technical deep dive into GSD 2.0's architecture: anchor pruning, deterministic tools, fractal summaries, and the engineering that makes AI agents reliable.
  20. Spec-Driven Development: GSD vs Spec Kit vs OpenSpecA practitioner's comparison of the three leading spec-driven development frameworks for AI coding agents. Architecture, trade-offs, and honest recommendations.
  21. Claude Code Agents and Skills: A Practical GuideLearn how Claude Code agents and skills work, how to create custom workflows, and best practices for getting the most out of AI-assisted development.
  22. Spec-Driven Development with GSD: A Complete GuideSpec-Driven Development turns AI coding from chaotic prompting into systematic engineering. Learn how GSD works step by step with real examples.