writing

thoughts on applied AI, infrastructure, and building real systems

featured

apps are the flywheel, data is the asset — building a portfolio of AI apps? bet on the data

the three-party identity problem in mcp servers — the architectural challenge every agent system hits

is claude code secure? — secrets, prompt injection, and the real weak link

all posts

• your ai agents are acting on behalf of your users
  and your backend systems have no idea who they are
• the governance controls existed
  they lived inside the system that failed
• openclaw just shipped 40+ security fixes. here's what changed.
  here's what they patched, why it matters, and what it means for teams deploying AI agents in production.
• i scanned 20+ security tools on ClawHub
  here's what each one does, where they overlap, and which ones to actually install
• we scanned every skill on clawhub. here's what we found.
  a static analysis of 7,522 AI agent skills using regex pattern matching
• what tools is your openclaw agent using?
  gatewaystack governance — identity, scope, rate limiting, injection detection, and audit logging for every tool call
• someone forked my AI governance repo to distribute malware
  a real supply chain attack on an open-source AI security project, and what it says about trust
• don't trust the parrot
  the three-party problem in agentic AI, explained with a bird
• apps are the flywheel, data is the asset
  building a portfolio of AI apps? bet on the data
• i automated my build logs with AI
  how i use gemini to turn claude code transcripts into daily build logs
• is claude code secure?
  first impressions on secrets, prompt injection, and the real weak link
• how ai exposes poorly defined human systems
  automate informal systems at your peril
• what does 'relevant context' really mean for LLM applications?
  want to know? learn from a human
• user identity isn't a first-order AI system design principle
  and why this is as a fundamental architectural mistake
• resourcing genai initiatives
  don't resource genAI projects. resource learning
• prioritizing genai initiatives
  portfolio theory for genai
• sourcing genai initiatives
  practical advice to fill the top of your genai funnel
• most genAI initiatives never make it from pilot to production
  and why that's expected and okay
• does your team use llms securely?
  why chatgpt enterprise alone isn't sufficient
• how is an llm like a financial advisor?
  mcp servers are three party systems
• 1-line jwt/jwks verification for mcp backends
  npm i identifiabl — deploying gatewaystack's first layer into production
• the three-party identity problem in mcp servers
  thoughts around an agentic control plane for ai model access
• apps sdk vs mcp vs normal apis
  my mental model for understanding the emerging AI app ecosystem
• building an AI backend for early-stage products
  how i think about moving from idea to architecture to production for modern early-stage AI systems
• validating AI workflows with no-code mvps
  a simple approach to proving a problem exists before investing in engineering