Spotlighting Jinba: Turning Natural Language into Enterprise Workflows の全文書き起こし

but for them, like when it comes to ship it to production environment, like, for example, they have to provide the functionality to like, you know, tens of thousands of users, like employees, and they need to think about, permissions, like who can use this tool, who can access these files, those kind of stuff, where it's very complicated. And like for them, like they kind of realized, this is like not cost efficient. I mean, we build a prototype, but we're not going to ship it. So that was like kind of happening in a lot of like, you know, regulated enterprises. And yeah, so that's probably the biggest problem that they have. Samantha Keya: So what are the biggest bottlenecks? Engineering resources, cost, clients, Shoya Matsumori: Mm-hmm. Mm-hmm. I think one of the biggest problems... Yeah, it could be engineering resources. Yeah, those large enterprises, they're always saying, we're kind of lacking internal resources because they already have a big internal system, like ERP and transaction systems and those kind of stuffs. And when it comes to like start like new like AI automation project, they always like lack resources. So I often see people like, you know, in the, in the, in the bank, they're like outsourcing to system integrators, right. And then they, ask them to like build something for them, but it's very. Actually like it's costed and efficient because like system integrators, mean, some of them are like knowledgeable about banking industry, but like, you know, you need to do like a lot. of knowledge transfer and it's obviously like not really cost efficient. Samantha Keya: Yeah, for sure. Okay, so moving on to your product. Genva allows teams to build workflows by just chatting. So how does that actually work? Shoya Matsumori: Mm-hmm. Yeah. So essentially, like, you can think about, like, NA and it's lovable. So you can say, like, I want to, for example, every time I finish the meeting recording, transcribe everything, and then, like, send a summary to Slack. If you type something in, like, and then, like, you hit enter and wait for a while, or... AI agent would generate the YAML manifest. So YAML manifest is like the definition of our workflow. It's kind of similar to GitHub actions. And you can also see the graph representations of the... You know, it will have to work flow. And interesting part is it's kind of two sides of coins. And whenever you edit something on the YAML file, you can see the changes on the graph representation. You can also click on the node and you can like, you know, there is like a model and you can select the parameters like with GUI. And if you change something, you'll also apply it to YAML file. So yeah. Samantha Keya: how is this different from traditional no code and drag and drop tools? Shoya Matsumori: Mm-hmm. Mm-hmm. Traditional NoCo local tools, you actually need to figure out the settings by yourself. Imagine all this workflow, the output of LLM. is not very polished. So I want to change the model to

the latest model. But the thing is, you need to click the node, and you need to figure out the settings by yourself. And you just need to dive into very detail. So user actually has to have a mental model of those settings, which settings belong to which panel, like X, Y, Z. so in our platform, you don't really have to think about that. Like just ask, Hey, like I want to change this to like latest, you know, version of the model and like AI will just like go through the settings like automatically, you know, change the settings. Yes. And that's actually like a very, you know, huge jump. Samantha Keya: So I'm understanding is, is that Shoya Matsumori: Mm-hmm. Samantha Keya: you can be even less technical than before. Shoya Matsumori: less technical than before. And also we're like more on the vibe coding stack. So it's something similar to code code. Users can... Like ask, Hey, like I want to build this in our AI agent would run the workflow automatically and it will debug workflow automatically. So. I mean, also the, the, the hardest part, one of the hardest parts of like building workflow is actually like handling errors And nowadays, like, you know, if you're writing Python code by cloud code, it will just like iterate, like improve by itself. Right. So we also have similar kind feature for our platform and which is quite quite like powerful. Samantha Keya: How would you differentiate yourself from Cloud Code? Shoya Matsumori: Yes. So I think the important part is like security and guardrails. When you think about like code code, I mean, it's obviously for like, you know, technical user, but it's like super powerful and they can do like whatever they want. So a lot of people think like, why don't just like people in enterprise like use code code for development, right? Like it's way easier and it's also like, you know, it's very powerful. Some part I agree, like, you know, I mean, for technical users, like, you know, me, like I can just build something by myself, like, but the reality is like, if you, work at like bank or like insurance industry, you cannot Vive code something and you cannot like ship it to production one stop. You really need to think about like if the, you know, the Vive coded like implementation is secure or not. Like, is it actually working well? Is it actually meeting expectations like data and compliance and whatnot? So we... intentionally implemented some sort of limitations over what users can actually vibe code on our platform. For example, I explained we're using YAML file, right? So it's not technically a Python code or TypeScript code. it's... essentially like a combination of pre-existing modules. So it's not something you can write from scratch. And we intentionally gave a limitation because if people can vibe code something from scratch, it's very hard for other people to verify if it is safe or not. So YAML file is one of the growth drills we have. And we also have... You know, we can publish our workflow, but we also have like a fine grain permission settings. So we can like, integrate with their internal AD system and then like figure out like whether this workflow can be shipped to like department X or department Y, something like that. So if you use cloud code, you also need to do it by yourself. So that's kind of like, you know, like differentiator, huge differentiator between cloud code and us. Samantha Keya: What's the most interesting way you've seen someone use Jinba? Shoya Matsumori: one of the interesting use cases that I saw was, so one of our customers, she was working on the investment research and she was like essentially like creating a report in like PowerPoint format. she had to like create like a couple of like slides for like, you know, startups and like doing research. Like when you say something like, oh, what's the like startups, like what are the startups that raised like past couple of weeks in biotech industry or whatever. So she automated that workflow using Jinba. And I was specifically surprised because the output quality was very good, especially a slide, the design was very good. It was super polished. And I was surprised because that was made by combination of like cloud, like Opus 4.7 and also like Python code. it's like previously, like when you create like pitch decks, like you have to rely on like gamma or like specific like API endpoints that are specialized in like PowerPoint generation. But nowadays, like most of stuff can actually could be built with the combination of existing modules and also integrations. So I feel like I kind of see the potential of our product. You can create very complex stuff with a very simple combination of the modules. Samantha Keya: So one of the things that I've been hearing most in this conversation is the emphasis on Shoya Matsumori: Mm. Samantha Keya: clients and guard rails and how that built into Jinba. So Shoya Matsumori: Mm-hmm. Mm-hmm. Samantha Keya: how do you think about balancing speed of building with compliance and government? Shoya Matsumori: Right. That's honestly we're figuring out right now. And the current speed of technological advancement is very, very fast. It's super fast. I think even for startup, you kind of need to figure out whether to adopt... like this library or that library because like every single week, like we see a lot of like new libraries coming, If we like adopt like, you know, a specific technology and then like wait for like a couple of months, like sometimes they will get outdated, like it will regret like, you know, adopting that like tech stack. So... The same thing will happen, also happen for like, you know, banking and those like regulated industry, if they build something like very quickly, like you actually like regret it because it's actually like, it's going to be outdated. So it's actually kind of good for banking industry to think about compliances and they spend a lot of time because the technology adoption is actually like, how do I

say it? Like it's like five, six months, like later than the, know, cutting edge like field. So doing like those like five to six month, like you can actually think about whether to actually adopt this type of technology. example, like MCPs, people are actually skeptical when it came out, right? But those industries actually have a room to think about whether to adopt those technologies. So it's not essentially a bad thing to get lag behind of those technologies Samantha Keya: So one of the things that Jinbub boasts is use cases like loan Shoya Matsumori: Mm-hmm. Samantha Keya: screening, KYC compliance checks. Shoya Matsumori: Mm-hmm. Mm-hmm. Samantha Keya: Why are these especially well suited for automation? Shoya Matsumori: Right. I think the reason is like there are lot of documents and a lot of like varieties that they need to check. I especially work with the commercial bank and I was like so surprised like there are like a lot of bankers like well paid, they're like super talented but they had to work on like very weird like job, it's like checking numbers one by one, just copying like, you know, numbers to different Excel sheet one by one, like copy and paste. And I was like, and I talked to like those people and those people are like really smart and like, it's kind of crazy to see like people like working on that kind of stuff. yeah, but, It kind of makes sense because banking industry is like such a regulated industry. They're like Their entire value is like trust, right? So they are like super process focused. They want to follow every single right process, and they don't want to mess up thing. The process is well documented. They have like a certain procedure and they have very detailed documentation for that. So it's actually very suitable for like AI agents and AI because they're already like a rule in like natural language, right? So we can essentially like, you know, collect those like rules and then like we can turn it into like workflows. can like utilize AI agents to like actually follow the rules. So yeah, that's, that's the reason why like those areas is like very like suitable for AI automation. Samantha Keya: What happens to teams when these repetitive processes are automated? Shoya Matsumori: Right. I think the benefit is they can work on more creative stuff. Since a lot of their work have been distracted by paperwork and those kind of like... I won't say it's less important, but it's not for people who are talented like that. But once they don't really have to work for those tasks, can focus on, I don't know, probably talking to new customers, figuring out new strategy for upselling your product. Essentially, you can focus on more future stuff, right? So I think that's the huge benefit you can gain from automation. Samantha Keya: Why is explainability so important in enterprise AI workflows? Shoya Matsumori: When you think about running AI workflows, it has to be very predictable. And if the workflow is not deterministic

every single time, AI agent will decide by itself. But sometimes it leads to different direction or different conclusion. It's very hard to predict because essentially LLM is stochastic machine. they're like kind of looking for predictability in this industry, Samantha Keya: So Jinbo is already being used by tens of thousands of enterprise users. What's Shoya Matsumori: Mm-hmm. Samantha Keya: driving that adoption? Shoya Matsumori: think, first of all, it's very easy to build. whenever you actually want something, you can just build it ship it pretty instantly. It's like a one-click. And secondly is... We have a lot of integrations for those financial industries. like loan assessment, if they want to build something in loan assessment, like we have templates, so they can just ship it instantly. versus other platforms, like people have to build from scratch. So it takes time to actually like building something useful. when it comes to like automation, like you always think about the cost of building an actual like outcome, right? If the cost of building is like, you know, the smaller than like, the outcome. you want to build it, right? So you constantly need to feel, oh, this is easy to build. We can ship it instantly. And the outcome is very huge. Users want to use it a lot. But previous NoCo local tools are actually different. People think that, this tool, I can build whatever I want. And if you try it, you realize, oh, this is very hard. I can't use it. And people just stop using it. So I think that part is very, very different in our platform. And that's the reason why people use it you know, over and over. Samantha Keya: What's the long-term vision for Jinva as a company? Shoya Matsumori: Yeah, so we are essentially aiming to be an operating system for financial industries. So whenever people want to build something, we have Jinba and whenever people want to invoke AI agents, we have Jinba And we're just like keep building those useful tools for banking and insurance. Samantha Keya: So do you see a future where most internal processes are built and maintained without engineers? Shoya Matsumori: Yes and partially no. I think Samantha Keya: Okay. Shoya Matsumori: most of the crucial work still engineers have to maintain and build it. For example, like connectors, database access, those part are critical. However, most of the business process automation part, I think it's going to be like rest of reliant on engineers because business people working on you know that project or like yeah that specific work know a lot about the bottlenecks and like processes and maybe like hidden rules so they can like explain problem of their efficiency and they can also evaluate the result. So it's way faster if the implementation part could be done by others, like AI agents, and they can just explain something and build something by themselves. Samantha Keya: So right now, the traditional cycle of building takes weeks, months. Shoya Matsumori: Mm-hmm. Samantha Keya: What is your goal for the timeline that workflows take? Shoya Matsumori: like to build workflows, right? Right. I think right now like it takes not right now, but maybe like previously like even half a year ago it took three to four months to actually build and Ship something to production environment. I think even more maybe like six months in the bank But we were aiming two weeks, ideally. There are a lot of stuff that we need to take care of because when it comes to shipping workflow to production environment, you need to check like if the workflow is like secure, compliant and those kind of like reviews. Right now, in most of companies, people are actually checking one by one. But we can streamline the process itself by using AI AI will just review the workflow. assess the risk behind the workflow and people can just like take a look at it and approve or disapprove. So those type of stuff, we already have it and we're streaming the process. So eventually the cost of building workflows and the time that you need to spend building workflows will be very short. So after YC, we started open a of positions. So we have like founding engineer, we have founding designer, we have four deploy engineers. It's actually on our landing page, there is a career page you can take a look. And also we have YC job posting page as well. So yeah, if someone is interested in working like early stage startup, like vibe coding tool for workflow automation, yeah, please contact us. Samantha Keya: All right, you're here first. Thank you. This was such a fascinating conversation. Shoya Matsumori: Yep. Thank you so much. Samantha Keya: I can't wait to see how the platform evolves as more organizations adopt AI-driven workflows. And to Shoya Matsumori: Mm-hmm. Mm-hmm. Samantha Keya: everyone listening, we'll be back soon with more founders, more products, and more behind-the-scenes conversations. OK, well, thank you for tuning into 8 Spotlight. And thank you, Sholia. Shoya Matsumori: Thank you, Samantha.