Short answer: Rob Williams says teams should build a verb, not a noun: a crank that updates the product to new model abilities. Read AI avoided heavy prompt engineering in 2022-2023, expecting model improvements to unlock value.
Watch this answer from Episode 54 (starts at 12:19) · Episode 54 notes and transcript
From Episode 54: Nobody Takes Notes Anymore. 5 Million People Let AI Do It Instead., starting at 12:19.
Jordan Metzner [12:19]: While others talk about AI Ryz Labs is already implementing it. Got one last question. We discuss almost every week just like how fast the market's moving. There's new models from, like, all the foundational labs on almost a weekly, biweekly basis at this point. If it's not one lab, it's the other. Where did... Where do you see the future development of, I don't wanna say, like, AI or LMs or SI or whatever we're calling it now, but and the impact to your product over time as obviously these things get, more powerful, maybe cheaper, faster. I'm sure we could think of more adjectives, but, yeah, where do you kinda see it going and the impact on your product?
Rob Williams [13:03]: I think that the way we build is different, and it's it's a hard thing to, put your hands on necessarily when having these brief conversations with people. But what I mean by that is if you go back thirty years when we were building packaged software and physically shipping it to, like, an Office Depot shelf where it's shrink wrapped and available and it's static, you're producing... Your organization really is producing a noun that's just static. It's so important nowadays when you're building a product that's leveraging these tools to make sure that you're building a verb. Like the organization should be building the crank that we can turn to update the entire product line to take advantage of new abilities quickly as possible. Rather than just assuming what we have today is the steady state, it needs to be dynamically formed around the abilities of the underlying models. At the same time, a lot of things don't change, they just get better. And so if you design and build the product in the correct way, then you can almost abstract out all of that from your users and plug and play and drop in. But one of the early takes that we had in 2022, 2023 was that we didn't want to invest the same amount of energy that some companies did on prompt engineering back when that was all the rage, we assumed that that would be investment that was largely wasted. And instead, what we do is build the rest of the product in such a way that the assumed improvements would just, unlock value for free.
