Snap uses agents to find bugs and organize work around customer jobs

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

1. 45:46

   operates what are just like a few things you've that have been impactful and how your team operates that are ai driven taking a step back uh it's definitely not a requirement to ship code on the design team but i think people are so passionate and curious about learning new things i mean one of the things you know we've always tried to do on the design team is just figure it out right i remember in the early days of some of the 3d graphics stuff that we were trying to do one of our team members had never done 3d graphics before and they were like they're like okay let's just figure it out right so they took some online tutorials and like you know boom like we were off to the races so i think the design team has always been really flexible in terms of you know their curiosity learning new things and so i i don't think you have to like push people or create

2. 46:24

   a requirement or something like that for folks to want to adopt new tools and want to, you know, want to ship code. I think, you know, the big question now that everyone and anyone thinks they can, you know, submit a PR is like how we don't break things at a billion, you know, or nearly a billion user scale, right? That's really important. So I think what we've really tried to do is also develop AI tools and guardrails to make it easier for more people to, you know, participate and submit code, but also reduce the likelihood that, you know, we're going to have some sort of bug.

3. 46:53

   So, you know, we've got automated code review now. We've, I think, automatically detected like close to 10,000 bugs at this point, probably. You know, we've got, we have a system at Snap where, you know, on the internal version of the app, you can shake to report, you know, your problem and agents now, you know, debug exactly what happened, what went wrong and, you know, can actually like suggest a fix. And I think, you know, in pretty short order, it'll be implementing the fix as well, which is pretty crazy.

4. 47:19

   So I think, you know, taking a step back, there's so much excitement around the way that AI is going to change the way that companies work. Obviously, software companies in particular, like ours, I think, are on the cutting edge of the way that AI is changing how things get built and what gets done. We really wanted to bring some like order to that chaos. And so the way that we did that was really by starting with the jobs to be done for our community and our advertisers.

5. 47:45

   so getting really really basic and you know straightforward on okay you know for snapchatters right or jobs to be done it's as simple as like get people to download the app right get them to add their close friends you know get them to you know i don't know use lenses, something like that. On the advertiser front, it could be bringing people into the ad platform, configuring their campaign, et cetera. And so by listing out all these jobs to be done really for the community journey and for advertisers as well, it became very clear where we could use agents, where we needed to be very focused in terms of building cross-functional teams around those jobs supported by AI tools. And it's also given us, I think, a really helpful mechanism to track our progress against the business outcomes for each of those those jobs.

6. 48:31

   So I think, you know, for us trying to, you know, while I think in this moment of time, you certainly want to like, you know, have a thousand flowers bloom and people are building agents and experimenting. I think at the same time, making sure that that we stay focused on