What Does It Actually Take to Build AI That Works? Richard Socher Has Some Answers

‍Richard Socher, founder and CEO of You.com, sat down for a wide-ranging Q&A with Thought Economics on March 27, 2026. Read the full interview.

Richard Socher has been at the frontier of artificial intelligence longer than most—and has gotten there faster by being willing to be wrong in public. As a PhD student at Stanford, he championed neural networks for natural language processing at a time when the field was deeply skeptical. His early papers were rejected. His ideas were called far-fetched. Then Alec Radford and Ilya Sutskever cited his work in the early GPT papers, and the rest, as they say, is history.

Richard went on to invent widely-used word and contextual vector representations, pioneer the concept of prompt engineering, and serve as Chief Scientist at Salesforce after the company acquired his startup MetaMind. He then founded You.com—the first platform to integrate a large language model directly into a search engine—before the world had even heard of ChatGPT.

In a new interview with Thought Economics, Richard traces that intellectual journey in refreshingly direct terms. He explains why the leap from feature engineering to architecture engineering to prompt engineering wasn't a series of breakthroughs so much as a series of realizations that humans were still doing too much of the work. 

The interview also covers ground that's harder to find in most AI conversations: a genuine rethinking of the superintelligence safety debate. Richard argues that the dominant "P(doom)" framing contains a strange internal contradiction. It imagines an AI simultaneously intelligent enough to destroy humanity, but not intelligent enough to understand why that would be counterproductive. His alternative framing is worth reading carefully.

And then there's what he calls constructive optimism—the idea that technology, applied with agency, remains humanity's most reliable tool for progress. He draws a straight line from automated textile production to Uber to AI personal assistants, and makes the case that the goods and services currently available only to the very wealthy are exactly where you should look to understand where technology takes everyone next.

The full conversation covers the origin story of You.com, why enterprise accuracy matters more than consumer vibes, how to define intelligence as a multi-dimensional volume rather than a single threshold, and what the Luddites can teach us about managing the human cost of technological transitions.

It's one of the more intellectually honest AI interviews you'll read this year.