This week itās storytime with Steve Yegge! Steve came out of retirement to join Sourcegraph as Head of Engineering. Their next frontier is Cody, their AI coding assistant that answers code questions and writes code for you by reading your entire codebase and the code graph. But, we really spent a lot of time talking with Steve about his time at Amazon, Google, and Grab. Ok, itās storytime!
Matched from the episode's transcript š
Steve Yegge: Because we have to pay for GPUs for you, basically. So it was a bit of a logistics challenge to actually get Cody out to everybody. Next step is weāre going to make it an even more lovable experience. I mean, like I said before, a lot of your experience with any AI coding assistant, including Copilot, is going to come down to your tolerance for it making mistakesā¦ So weāre putting together an unbelievable world-class AI team, thatās pairing up with our code intelligence, our code graph, which is the thing weāve spent ten years building. Itās the external store. Remember I mentioned Google is going to be fine, because they have all this incredible search infrastructure that the LLM can use? Weāre in the same boat. Weāve got fantastic technology, so we can start improving Codyās quality.
Because ultimately - look, Cody is great right now if you have a question about your database, about some stuff you havenāt seen, itās new code you havenāt seenā¦ But for day to day, where you already know the code, and your IDE is helping you, and stuffā¦ And maybe Cody doesnāt come to mind as your very first stop, so my goal is, in very short order, we want to make it so that Cody comes to mind first when youāre thinking of a much broader class of problems. And we do that by making it really, really good.
You would be shocked at how much low-hanging fruit there is. I mean, you look at Cody and it does make a fair number of mistakes, but the ability for us to fix these things is extraordinary. Everything that weāve tried has made it better. So in other words, weāre starting with kind of the worst possible implementation of Cody. Itās OpenAI, generic, sort of Wikipedia embeddingsā¦ I donāt even think theyāre code-oriented, you know what I mean? So yeah, itās gonna go out better.
Where are we going with it? I mean, long-termā¦ Look, I want Cody to be your first stop when youāre coding. Look, one of the things that I was always jealous of, truly envious, about VPs with executive assistants, as I was going through my career, was that they were always like a team of two people. And they were always incredibly productive, and the VP never had to go to the printer, never had to do any of the expense reports and stuff that engineers are always having to do. The VPs were always freed up. I mean, honestly, thatās what Cody should be for a developer. Itās like āOh, man, Iāve got some stuff I need to get done, that I really need to get done that Iām struggling with. Let me just have the AI help with it.ā Honestly, thatās where I want to take it. And itās so early days right now. I mean, itās really raw right now. It adds a lot of value if you play with it. But still, I mean, you probably find that it also has a lot of mistakes. But the way I see it - I mean, this is inevitable. This is an inevitability. Itās going to happen. It may take a year, it may take five years before weāre all using it every 10 minutesā¦ But itās inevitable.