Warum das Warten auf das nächste KI-Modell sich lohnen kann – John Stecher, Blackstone & Scott Worland, Norm AI

설명

Blackstone CTO John Stecher and Norm AI CTO Scott Worland discuss building technology that holds up as models keep improving. John explains why a solid data strategy and straight-through processing come before agents, how he weighs ROI against waiting for a better model, and why internal apps can now be thrown away and rebuilt. Scott describes how Norm's large hand-labeled eval set showed that much of its carefully built workflow code was in the way, why the team rebuilt its platform, and how it decides where to invest heavily versus lightly as the frontier moves. Chapters: 0:00 Intro 1:05 Getting AI agents right 1:32 Data strategy comes first 2:21 Straight-through processing still matters 3:46 Norm's early workflow approach 5:05 When agents made the code obsolete 6:42 Evals in financial services 7:40 The bitter lesson 8:54 ROI, and when to wait for the next model 10:25 Getting to a proof of concept fast 12:43 Updating on Claude Code 14:49 The hard part: getting to production 15:45 Parts that still need human review 17:24 Rebuild instead of maintaining 19:13 Closing thoughts John Stecher https://www.blackstone.com Scott Worland https://www.norm.ai Join us in SF for the second annual AIE CODE summit: https://aie.codes our last flagship summit of 2026. Applications live now!