Obi is Back
Thoroughly Considered
The episode explores how small-scale technological implementations impact user experience and business efficiency, starting with Studio Neat
Key takeaways
- Studio Neat developed 'Obi,' an internal AI operations bot that uses Claude API and Markdown files as a legible knowledge base.
- The 'Obi' system connects to various business data sources like Airtable and Shopify to facilitate inventory planning and ad analysis.
Main topics
- Obi AI tool
- Studio Neat operations
Notable quotes
What a good interface chatting is.
Conclusion
A fascinating look at how even simple technological implementations—fr
Transcript preview
Speaker 3 (0:07) Tom Gerhart and Dan Provost are the guys behind Studio Neat. On this show, we discuss entrepreneurship, product design, and all of the ups and downs that come with running a small business. I'm Mike Hurley, and this is Thoroughly Considered. Speaker 3 (0:31) I've got to say, I opened our Apple Note. There's Speaker 2 (0:33) a Speaker 3 (0:33) bunch of stuff going on in here. Dan, I feel like you've got some topics for us today. You've got a whole list. You've got a whole bulleted list of stuff here. I've got to say, I'm very intrigued by the top one. This is Obi. Is it back? Obi's back. Is it back? Obi's back, baby. Speaker 2 (0:47) Oh, it's Speaker 3 (0:47) back, Speaker 2 (0:47) baby. Speaker 3 (0:49) One more. One last ride. Speaker 2 (0:52) That's right. One last ride. Yeah. Why, Dan, how did, is there some, was there some other reason why we chose that name for this new thing? Or did you just joke and say. It Speaker 1 (1:06) just came to me and it was kind of a backwards thing where it's like the letters OB could be like operations bot. And then it's like, oh, OB, like we have that name already. Let's use that. Yes. Yes. So what are we doing, Tom? What is it? Oh, wait, hang on a minute. i Speaker 3 (1:24) was not fully following so there is there's something new called obi yeah Speaker 1 (1:30) but it's it's Speaker 3 (1:31) it's Speaker 1 (1:32) internal only no one else will know or care about it okay Speaker 2 (1:35) we're making a like ai uh i guess like operations like helper thing at studio neat right and we just are calling it obi that's brilliant and uh And yeah, it's basically... Speaker 3 (1:49) Let's hope Speaker 2 (1:49) that's not a Speaker 3 (1:50) curse. No, Speaker 2 (1:51) no. Hopefully not. But yeah, so basically it's like we're trying to... Not really make a typical AI agent. More of a place where we can just use natural language to ask questions. And then it's connected into all of our data sources. Oh, yeah. And it can just answer those questions. And then also, we're starting to use it because now it's easy to vibe code. a way to get all that data in one place now we can use it for doing some of this planning stuff that we do a pretty good job of but we miss little things or it's hard to think through all the downstream consequences and so this is Speaker 3 (2:40) the saspocalypse stuff right of like Speaker 2 (2:42) um no but it's like an example right yeah Speaker 3 (2:45) because i'm seeing this but everyone i know is is building some kind of tool like this right where it's like yeah you have a bunch of data sources available to you as part of your business. They have APIs. You can connect to them. And there are tools that can try and bring them together, but you then have to work their way. Yeah, yeah, yeah. Where this is like, you know exactly what you need as a business. You have the specific set of tools that you use, so you're able to build. a tool that brings all of that data together and that you can have a conversation with. Like it's a, I think it's a very natural thing for a lot of businesses to have now. Speaker 2 (3:19) It totally. I mean, it's funny though, because we're still using Airtable and I think we'll keep using Airtable, which is the normal, it's like the SaaS tool we're paying for, right? And that's like, basically, it's like a database. but with an interface that's like integrated. And so we still want to use that interface. Like I think we could vibe code our way into that, but it's like something you have to maintain and we already know the thing. So this OB layer is more of a like intelligence analysis thing. And then it can propose things that we then like review and then write into our like systems of record, quote unquote, like Airtable. Like what kinds of things Speaker 3 (3:58) would it? could it Speaker 2 (3:59) it's mostly just like inventory planning uh like when we need to order new parts um that's most of the stuff we're gonna start doing it for like uh ad like creative and ad analysis too like like because we have all this stuff and we can see like we get ad reports from our like from our advertising team like our media buyers advertising team but we that data is not like not exactly integrated with like say the Shopify data all the way and so there's some advantages of like collapsing everything yeah but I think the thing to maybe talk about that's interesting and I don't know Dan why you even put this on here if you have some other reason but from my perspective I think one cool thing to put on there or to uh to say is the way we're designing it is like really uh simple and cool in a way um Basically, I just was chatting with Claude like, okay, hey, we want to do something that gets us these goals. How can we do this in a really simple way? And what it ended up being is it's basically a series of eight markdown documents. And what I do is I'll go in and I spend a couple 45-minute sessions just doing one long dictation of how the business works. What are all the tools? Who, like, what are the people? What is the history? Like, what are the values of the company? Just kind of like trying to dump, like, in a long rambling thing while I was driving, like, everything. And then Claude takes that and compresses it into, like, a formatted, like, almost like a knowledge base. It's almost like a wiki document of, like, this is Studio Neat. Even, like, heuristics. Like, if we say, hey, Mark 1. It knows this SKU, like it has the colloquialisms of the business, right? Like we can just kind of speak in casual ways. But the cool thing is that's just a markdown document. And then there's like some config files that it's like, here's all the data sources. Here's how they work. Here's like a description of like all the air table and how it works. And so it has the connectors into those things. And then just these documents that are like the reference documents. And then we're basically just using Claude, like normal Claude API. to do the intelligence piece but the really cool thing about it is all of the like what obi knows about studio neat is like incredibly legible because it's literally just like in this document so there's we'll have sessions where we'll just like we haven't done this yet but the idea is basically you just like edit that document either verbally or whatever this Speaker 3 (6:23) is like harness stuff yeah yeah exactly people building personal harnesses this is the thing that david sparks has been doing a lot and it's where i learned a bit about it and he He's been showing me some of this and he's built courses around it now of like basically like you can just create a bunch of documents and you just keep having the AI write to the documents and update the documents, which also makes it incredibly portable as a system. Yeah, that's Speaker 2 (6:44) what's nice about it and legible. It's like you know what it's like. It's not this large logic chain of what's going on. It's literally just like here's what the rules are. They're right here. They're really simple. So that just feels like... Like, we're not getting in this deep, like, rabbit hole of, like, all this stuff. And, like, explicitly from the beginning, we're very conservative. So, like, the whole point is really just for us to see data. And then if there's any going to be any right or change to anything, it's, like, about confirmation. It's like, we need to see what you're going to write and, like, prove that it's accurate. Like, it's all built on being able to, like, kind of show proof that that's the right thing. So, it's just, I don't know. I was just really struck by... And I've been kind of casually working on this for a couple months. And just now it's starting to turn the corner where it feels like, oh, this is valuable. And I think it's going to be more and more valuable. But what feels cool to me about it is... it's just flexible enough like it's it's flexible enough to where we could design this system in a studio neat kind of way where it's like simple and straightforward and it's not this huge like thing that's opaque and all this stuff and so i don't know i think it's just it's like cool it's like a cool um cool little situation i Speaker 3 (7:58) think it is a i think it's very strange right where like these tools are being built right now Speaker 3 (8:07) in these ways that are very simplistic. Like a lot of the really powerful stuff that's happening on Macs right now is happening in Markdown documents that systems read and write to and accessibility hacks that they're able to work around. And it's like, this is how the really smart machines are working. Like they're just... They're just using such basic Mac tools because the system is open enough to allow them to do it. And it's just a very fascinating thing that's happened, which is almost like it's born of constraints of the platforms, right? The Mac is the only computer that will allow you to do things in this way, but it will also only allow you to do them in very specific ways. the systems have to just work around all of that. It's a very strange time. Speaker 2 (8:56) It kind of feels like iOS days too, especially early iOS days where those, remember the like, what was that initial like David Bernard? Launch Center Pro? Launch Center, yeah. It felt like a launch center timer. It's like, you know, it's like, yeah, there's these hacks. They're very specific and then you can use them artfully to make this thing. It feels kind of like that or something. Speaker 3 (9:19) Yep. Speaker 2 (9:20) Yeah. Speaker 2 (9:24) Um, dad, did you have anything else that you wanted to say about it? No, no, just Speaker 1 (9:28) that. I mean, it's, it's a testament to. Speaker 1 (9:34) This is like a very obvious thing that's been said many times before, but just like, what a good interface chatting is. Yeah, yeah, Speaker 2 (9:47) yeah. Speaker 1 (9:47) It's like really, ever since this AI stuff started coming on and started using ChatGPT or whatever, it's like, this is really... The first time where I really felt kind of Steve Jobs is like bicycle for the mind idea where it's like previously a computer. There's there was always this level of like, you have to know how to use a computer to to do it. And this is just like. there's something there's like a different level or something with the interaction where you're just talking a natural language and receiving answers back in natural language where it's like whoa this is it cannot Speaker 3 (10:29) be overstated for a small business owner how impactful it is to be able to ask questions to a Speaker 1 (10:37) data set yeah yeah Speaker 3 (10:38) yeah yeah you can have your data of your sales and you can ask the computer what does this mean? What does that mean? But why did this happen? And it can give you answers. Like it's an unbelievable system. Cause otherwise it's just