Field Notes
Things I've figured out in production — written down so you don't have to figure them out the hard way.
Founder Decisions
Cursor vs Claude Code: A Founder's Pricing Teardown
Cursor pays Anthropic $650M/year while earning $500M. I break down pricing, trust, business model risk, and why I chose the $100/month option with no middleman.
How to Build Your Startup's Website in 2026
WordPress, Webflow, Next.js, or Astro? A real comparison, and how this site was actually built.
Engineering
One Million CCTV Events a Day in a 512 MB Container
A million camera events a day is only twelve a second. The real work is keeping one fragile NVR stream alive on the smallest container on a shared box. Why Go and the actor model fit the shape of the problem.
Elasticsearch in Production Is Not a YouTube Tutorial
Every YouTube tutorial makes Elasticsearch look easy. Then you run it in production. A node pegged its CPU and stopped answering queries, holding just 4.2 GB at 39% heap. It hadn't failed. The setup around it had: a schema never shaped for its queries, and storage that was never bounded. Here's what the tutorials leave out.
I Built an AI Agent That Runs Untrusted Code to Find Security Threats
877,000 malicious packages and counting. I used Upstash Box to build a sandboxed AI security scanner that does both static and dynamic analysis.
GitNexus Benchmark: Graph-Indexed Retrieval vs Claude Code's Native Exploration
Does graph-indexed retrieval actually beat Claude Code's native codebase exploration? I ran the same architectural question both ways. $0.52 vs $0.39, 38 tool calls vs 12, and the KuzuDB internals that explain the difference.
Work-Stealing from First Principles
Tokio, Go, and Java's ForkJoinPool all use work-stealing. I learned the algorithm by implementing the Chase-Lev variant from scratch in Rust, building up from load distribution to per-worker deques, CAS, and memory ordering.
Not sure what to build or how to build it?
That's exactly what the advisory session is for.
Book an Advisory Session