Tassenger is the clearest agent-directed delivery proof: I rebuilt v3 across native iOS, native Android, a web console and the Supabase backend in about twenty days by directing AI coding agents through 21 requirement rounds, 9 architecture decisions, 58 SQL test suites and 230 UAT documents. The speed matters; the verification system is the real signal.
Mastroify proves a different operating model: an autonomous content engine that combines real astronomy, LLM-directed writing, local text-to-speech, Playwright, ffmpeg and native publishing APIs. It produced 472+ videos over about seven weeks, operated across 11 platforms and cost about one dollar a month before I paused the studio after the model was proven.
The data work is operational, not ornamental: Aircraft Flow 360° surfaced delay causes and predicted turnaround risk at Changi; FMT's Chartbeat-to-MongoDB pipeline powers trending and related-content signals; and MXP gave marketing teams a self-service experimentation platform. I turn those systems into analytics reporting senior management can act on.
Automation at FMT follows the same rule. The YouTube-native Video CMS replaced polling with WebSub, reducing API quota use by about 80% and update latency to roughly 30 seconds. The useful result was not a clever integration — it was a simpler production system with fewer moving parts and clear operational controls.
What's here is verified agentic delivery and data systems in production — receipts, not demos — and I'm straight about where my edges are.
