Why is it so hard to get ROI from AI? Because building from first principles isn’t easy

Hello and welcome to Eye on AI. In this edition…highlights from Fortune Brainstorm Tech…Anthropic walks back a controversial decision around its new Fable model…OpenAI considers slashing prices…Meta looks to boost subscription revenue…and Anthropic’s Dario Amodei has a new AI policy manifesto.

It’s Jeremy here. First, I want to thank Sharon Goldman for her work as my co-writer on this newsletter over the past two years. Sharon, who normally writes the Thursday send of Eye on AI, is going off to become her own one woman media empire. We wish her all the best in her new venture.

I spent the first part of this week at Fortune Brainstorm Tech in Aspen, Colorado, where AI dominated every conversation, on stage and off. On Wednesday morning, I even hosted Fortune’s first Eye on AI breakfast, where it was great to meet some readers in the flesh. We discussed the reasons many companies are struggling to realize return on investment from AI.

Principles before profit

Manoj Bohra, the CTO at asset management firm State Street, said that part of the problem was the amount of what he called “foundation work” that companies needed to do to enable AI projects to succeed, especially in regulated industries. Having the right data in the right places with the right governance ...