In this episode of the AI + a16z podcast, Tony Holdstock-Brown, co-founder and CEO of Inngest, joins Yoko Li, partner at a16z, to discuss the realities and complexities of running AI agents and other multi-step AI workflows in production. Tony also explains why generative AI development tools and their founders are so similar to previous generations of these products, and where there is room for optimization.
Below is a sample of the discussion, where Tony shares his advice for engineers looking to build AI.
“As engineers, we have almost two parallel tracks right now. On the CPU track, everyone's like, 'Oh, it's CPU-bound, it's big O notation. What are you doing on the application-level side?' And then on the GPU side, people are doing crazy things to speed up numbers, to make differentiation better and smoother, to do gradient descent in a better and more powerful way. Right now, the two fields are working together, but they're very, very, very different from an engineering perspective.
“That's an interesting piece to think about for new engineers, or people who want to go into engineering in general: Do you want to be on the AI side, taking all these models, running all this stuff, building application-level stuff, chaining different things together to build a product? Or do you want to be on the math side, doing the very low-level stuff to make compilers work better to make the AI run faster and more efficiently? They're both engineering, but they're very different applications.”
