In September 2026 I shared a series of LinkedIn updates covering a quick reference on inference engineering, a partnership with Convex for the Build Sprint Season 04, the launch of a two‑week Build Sprint program, research on ultra‑low‑latency voice AI (Tap.Szn 03), and observations on AI‑related career dynamics and the ChatGPT Ads business.
Welcoming Convex as the Database & Hosting partner for Build Sprint Season 04.
We've been using Convex ourselves to build the GrowthX community platform, once you use convex you cannot go back.
The goal was 500ms for TTFA.
Cost had to stay under ₹2 a minute.
People quoted ₹1–3 a minute for systems running today, with ₹1.5 as the number to beat.
network: 50 ms
STT: 200–300 ms
LLM: 0.3–1.5 s
ChatGPT Ads business recently crossed a $1 billion annualized run rate.