writing · 76 essays
Essays on engineering leadership, the agentic SDLC, platform engineering, and enterprise AI — what actually constrains software delivery.
// featured · August 11, 2026
I automated audio versions of my essays, looked away for four months, and found a 22-episode back catalog. Meet Never the Stack, the podcast I never meant to start.
Read essay$ vinny init agentic-sdlc → spinning up reasoning loop ✓ spec generated · 217 lines ✓ tests scaffolded · 14 cases ⚠ coverage gap detected · branch /auth → prompting refinement ✓ branch coverage · 96% // shipping behavior, not just code
AI can create a real time dividend for engineers, and history says unallocated capacity gets absorbed before anyone notices. The case for deciding where the saved hour goes.
Open-weight models have raised the local capability floor while hosted intelligence keeps getting cheaper. The strategic question is no longer which side wins, but which model earns each workload.
Entry-level work was never just cheap execution. It was the training system that produced the people capable of judging what the machines build.
The rewrite taboo broke for systems that can grade their own replacement. At enterprise scale, the first modernization dollar should fund the answer key, not the port.
AI is making execution abundant. As intelligence becomes cheaper and more accessible, durable organizational advantage shifts toward judgment, context, verification, and leadership.
For twenty-six years, the full rewrite was the cardinal sin of software engineering. Then one engineer ported half a million lines of Bun from Zig to Rust in eleven days. The taboo was never about code. It was a price tag, and for systems whose intent lives in an executable oracle, the price just collapsed.
Satya Nadella named a real problem: AI workflows generate valuable learning exhaust from proprietary context and corrections. He is right about the asset. My practitioner's edit is simple: own the learning loop, not necessarily the model.
My personal delivery pipeline runs on three Claude agents, three nested loops, and two explicit human approval gates. The gates sit where judgment is cheapest and most consequential, and the real bottleneck moved upstream to the clarity of the ask.
In June I itemized the AI subsidy by hand. This month I built the instrument that does it for me. It took an afternoon, and it was built by the very thing it measures.
I built a serverless internet health monitor because I wanted an excuse to learn Lambda MicroVMs. The better lesson was that Barometer did not need the newest primitive. It needed the boring, correct fit between the tool and the job.
featured