AI startups need workflow integration and data flywheels

Peter is discussing two key strategies for AI startups: data flywheels (systems that continuously improve through user data) and workflow integration into users' daily processes.

The data flywheel part is just, you can start with proprietary data, but the flywheel is really just sort of how do you continue to maintain that and generate that. And the second thing is, again, it's the workflow.

Foody is CEO of Mercor, which helps AI companies hire experts to evaluate and improve AI model performance.

The wealthiest companies in the world are willing to spend whatever it takes to improve model capabilities.

Harvey is an AI startup that builds specialized legal tools and custom models for lawyers and law firms.

We're deeply focused on these very, very general use cases like the general reasoning capabilities, the general coding, the general writing abilities. I think where you start to get into some of these very vertical applications... that's a great example of our models are probably never going to be as capable as some of the things that Harvey's doing.

People really underestimate where the value is created in these applications and they just kind of get it completely wrong. It's not the UI that matters and it's not the data model that matters, although those are both very useful. It's the years and years and years of evolution of the underlying workflows of the product to support the customers.

Mike Krieger is discussing Anthropic's AI models (like Claude) and the intersection between model development and product experience development.

If we're shipping things that could have been built by anybody just using our models off the shelf, there's great stuff to be built by using our models off the shelf by the way, don't get me wrong, but where we should play and what we can do uniquely should be stuff that's really at that magic intersection between the two.