AI Engineer World's Fair 2026 — unofficial import demo

Somebody in your company is going to ask what you're getting for all that AI spend. If you don't have a good answer, someone else will make one up... and it might be "total tokens consumed". That's how tokenmaxxing becomes policy: not because anyone thinks it's a good metric, but because engineering didn't offer a better story. I work with datasets spanning hundreds of companies, hundreds of thousands of engineers, and billions of lines of shipped code to understand how AI engineering is evolving and what actually matters to measure. One thing I've learned is that raw token spend is a VERY crude estimator of value. For example, across levels of token spend, cost per merged pull request varies 300x — but output only varies 2x. The good news is the data also shows what DOES matter, and it's measurable and actionable – but most teams aren't tracking it yet. This talk will give you the data, metrics, and frameworks you need to keep your org from adopting the latest terrible vanity metric. You'll learn what actually separates teams that scale AI effectively from those just burning tokens, and how to tell the story that keeps your AI investment funded and growing.