Inside OpenAI's Real-Time Access Engine

​Provably Correct, Impossibly Fast

​In February 2026, OpenAI published "Beyond Rate Limits" - a deep dive into how they rethought access control for Codex and Sora. Instead of choosing between rate limits and usage-based billing, they built a hybrid real-time system that makes a per-request decision: how much is allowed, and from where?

​They called it the "decision waterfall."

​In this session, Jonah Cohen - Tech Lead for Financial Engineering at OpenAI and the architect behind that system - joins Stigg for a candid look at what it took to build it, why off-the-shelf solutions fell short, and what he'd tell engineering teams facing the same problem today.

​We'll cover:

• ​Why "metering" and "decisioning" are fundamentally different problems

• ​How OpenAI fuses rate limits, credits, and entitlements in a single synchronous request path

• ​The architecture trade-offs behind provably correct billing at scale

• ​What this means for every AI company shipping usage-based products

​Whether you're building your own access engine or evaluating infrastructure to handle it, this is the most honest conversation you'll hear about what it takes to get real-time usage control right.