An internal agent can become a tool everyone improves

Transcript

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

1. 26:17

   maybe you can show us one more thing like i'm um you know like um cloud launched tags recently and people are saying like oh you know we're moving from the id to like the app and now we're moving to just like people giving commands to agents in slack and um like i think you guys have a bunch of internal agents right like there's like um yeah natural or something or like honeybot yeah Yeah, so we have this whole internal dashboard thing. So this is all kinds of features people built, dashboard, internal tools, all kinds of things. I'm really proud of this. I think um it's set up in a way where everyone on the team is adding stuff in here and building it and um you can see like we have this little edit with cursor do that like uh you can if i turn this on and i'm like um make this red um it like you can basically like quantically be like hey i need to

2. 27:20

   change the internal tool to go do this thing for me um oh nice okay gotta go spin up a cursor agent to go do it so we've really tried to like like a bunch of people have put in work to lower the bar to making internal tools yeah nacho is like i think by far and away the most used internal tool thingy we have um it's a it's a chatbot here um and it's high level but it's like connected to uh as many things as we can give it like slack granola amplitude all of our logs okay website analytics like as many things as we can as many things as we can and then people use it i guess you can use it here it's mostly used in slack um some examples that hopefully safe to Yeah.

3. 27:59

   Like look at the. Code, figure out how the code works, give a smart answer, yeah, you know, and like, this takes so long to go, actually dig something to yourself. It's like, so this was a minor, I think, summary generation had a little dip for a second or something, like a minor incident. And, yeah, we have like the incident now starts out, Toby asks what's going on. So this is like, what's this doing?

4. 28:22

   This is looking at logs, I think, to figure out, you know, where are these errors popping up, what's going on there. it's then able to then go like sense check that against the code and figure out like why would that error be happening i see uh yeah i'm like this this um natural basically gets brought into every like every question that pops up these days uh uh this is the first protocol and um yeah i don't know we're all extremely at this point basically you just linked it up to multiple internal company sources and has a you can it sounds like you can pick the model that you want to use but then like it kind of does the first pass response right and you can talk to it yeah Yeah. Yeah. And, you know, it's, it can, it's, it's really reliable at SQL queries and things like that. So it can pretty,

5. 29:03

   yeah, you can pull data in and it's a good shoot, like just, just way, way faster than a human D3 dashboards would be able to. Yeah. And do you have like some guy or some person working on that? So like an internal tools team? Nobody owns it. Like, I, like, um, uh, JK, JK one of our engineers, I think started it as a weekend or an evening project. Um, yeah. And no, people just add things here and there to it.

6. 29:27

   like um yeah someone like natura is incredibly popular and so if you make natura better you'll be everybody's best so uh there's a strong incentive to to you know if you if you've got 10 minutes to try and figure out a way to make natura more powerful for everybody yeah i i feel like this is the right way to do things man because i i feel like um like i guess you can also just bring clod or you know like a cursor in here too but like i feel like companies should just build their own internals thing for this it's not too hard to build i feel like and like more flexibility like if if someone's model becomes better or like the price gets too high or something you can just switch right yeah yeah and it's i mean we have um we have i think natural can create cursor uh agents so like it will often you know like in doing an investigation will often

7. 30:10

   lead into a pr being written to fix the thing and i think that's got way smoother since we added cursor and made that not two two different tools yeah so i guess uh like i guess just to recap like Like, you know, how do you be AI native and still give people a life is like, it's kind