How I built a zero-trust security layer for multi-agent AI pipelines in Python after a agent-privilege escalation incident took down our staging environment.
Building a TypeScript feedback sensor system that catches semantic bugs in AI-generated code through mutation testing, property-based testing, and AST analysis.
How I found prompt injection vulnerabilities in my own MCP servers and built a multi-layer defense system that actually holds up under adversarial testing.
How I built a typed spec framework in TypeScript that stops AI agents from building the wrong thing. Zod schemas, just-in-time context loading, and validation layers.
How I built a secure registry for AI agent skills using content-addressable storage, Sigstore signing, and sandboxed execution to prevent supply chain attacks.
How I built a TypeScript AST scanner to measure and eliminate cognitive debt - code that works but nobody on the team can explain.
How to build AI agents that survive crashes and resume without corrupting state. Event sourcing, idempotent tool calls, and audit logging in Python.
I've been using Claude Code with Figma's MCP server for a few months. Here's what genuinely changed, what cost me time, and where I'm actually impressed.
Real defenses against npm supply chain attacks, grounded in official documentation and one well-documented incident. Written by a senior security engineer who has actually had to respond to these.
How I built a TypeScript runtime interceptor to catch AI agents leaking API keys and secrets through commit messages, PR descriptions, and tool outputs.
Building a Python tool that combines deterministic dependency graph analysis with LLM-based evaluation to catch architecture violations before they ship.
A tactical guide to writing RFCs that align engineers, PMs, and designers without burning everyone's time in sync meetings.
Traditional productivity metrics fail for AI-assisted development. Here's a framework for measuring what actually matters when your team works with coding agents.
Beyond the hello world tutorials. Real patterns, common mistakes, and what I wish I knew before shipping FastAPI services to production.
Error tracking, logs, and traces without the enterprise complexity. What I use for production monitoring on real projects and why I stopped overthinking the stack.
How I used AI to understand how card pricing sites work, figured out their APIs, and built a custom Pokémon card investment tracker from scratch.
AI closed the technical gap between engineers. The real differentiator now is communication, honesty, ownership, and the ability to make your team better.
Traditional APM can't tell you why your AI agent made five model calls and picked the wrong tool on the third. Here's how structured tracing with OpenTelemetry changed how I monitor agents in production.
Writing great code is table stakes. Systems thinking is what lets you understand why your perfectly correct change broke something three services away.
I built an AI code review pipeline that queries Sentry before commenting on a PR. Here is the architecture, what it catches, and what it took to make it trustworthy.
GEO (Generative Engine Optimisation) is how you get AI platforms like ChatGPT, Perplexity, and Google AI Overviews to cite your site. A practical, developer-focused audit and fix guide.
The mindset shift from ticket-closer to product thinker: what changed when I started asking why before writing a single line of code.
Most SKILL.md files fail because of two lines at the top, not the instructions. Here is how the loading system works, why the description field is the trigger condition, and how to write skills that actually activate.
Cloudflare Browser Run gives AI agents a real browser on a global network - with Live View, Human in the Loop, CDP access, session recordings, and 4x higher concurrency. Here's why it changes how I build agents.
How I built a CI loop that covers the full stack - Cypress for backend behaviour and E2E flows, Chromatic for UI visual confidence - anchored by a small-branch culture that keeps feedback tight and deploys boring.
Six attack vectors specific to MCP server implementations - confused deputy, token passthrough, SSRF, session hijacking, local compromise, and scope inflation - and the concrete mitigations I apply before any server ships.
Learn how to install and configure cavemem to give Claude Code, Cursor, OpenCode, and other coding assistants cross-session persistent memory using compressed SQLite and MCP tools.
AI coding agents default to the shortest path, skipping specs, tests, and security reviews. Agent Skills fixes that by giving agents the same structured workflows senior engineers use in production.
A shared Button change silently broke 12 screens in production. Here's the testing stack - Cypress for behaviour, Chromatic for pixels - that ensures it never happens again.
Learn how to build a Model Context Protocol server in Python using FastMCP, expose custom tools to Claude, and connect it to Claude for Desktop via STDIO transport.
Claude Code re-sends your entire conversation on every message. Learn how CLAUDE.md, /compact, /commands, model selection, and MCP hygiene can cut your token spend by 60% or more.
Stop wasting engineer hours on review bottlenecks. Use OpenCode inside GitHub issues and pull requests to automate triage, inline fixes, and PR reviews - without leaving your workflow.
Node.js to Bun in the Frontend: A Safe Migration Path That Won't Break Your CI - 3-stage approach: package manager first, then tests, then runtime