16 Jan 2025 · 30m · Elad Gil, Jesse Zhang · 26 quoted facts
Customer support is the best early market for AI agents because the work is easy to measure and can be rolled out gradually. Decagon CEO Jesse Zhang argues that value in agent companies comes from the software and orchestration built around the same shared models, especially transparency tools that let customers see and control what the agent does.
The tension runs between the speed of agent adoption and its limits. Zhang expects agents everywhere in customer service, yet is bearish on most other uses in the near term. Agents work only where they can start imperfect and grow, and where return on investment is easy to quantify. Voice latency and instruction following remain open problems.
Support teams already track conversation volume, satisfaction and NPS, so buyers can test an agent on 1% of traffic and compare vendors on hard metrics. Zhang says this makes evaluation quantitative rather than speculative.
Everyone can access the same models, so Decagon builds orchestration that combines models around a customer's business logic and tools to inspect the data and steps the agent used. Large buyers do not want a black box.
Zhang says gains in coding and quantitative reasoning from models like o1 help only modestly. Support agents must follow a customer's workflow exactly, so he is waiting for research on instruction following.
Voice-to-voice models respond faster but leave little room to fetch data or make multiple model calls. Chaining speech-to-text, a language model and text-to-speech adds delay, which teams mask with phrases like 'give me a sec'.
Zhang says tasks needing near-perfect accuracy from day one, such as security monitoring, will be slow because models are non-deterministic. Copilot-style tools like text-to-SQL are hard to price because the return on investment is hard to quantify.
Jesse Zhang says the work suits large language models and is easy to measure, with metrics like share of conversations handled and satisfaction scores. Companies can also start with a small slice of volume and expand.
Decagon ships AI agents that handle customer service conversations for companies. Zhang says its focus is transparency, so customers can see what data and steps the agent used and give feedback.
Zhang says Bilt Rewards stopped scaling its support team within about a month of starting. Almost a year in, the savings equal roughly 65 agent headcount, according to a published case study.
Callers expect quick replies, but turning speech to text, running a model, and turning text back into speech adds delay. Faster voice-to-voice models help, but may not allow the extra lookups some customers need.
He points to areas like security monitoring, where mistakes are costly and models are non-deterministic (not always giving the same answer). He also names hard-to-price helpers like AI data scientists.
Stated by the speakers. Each quote matches the transcript, which shows what was said, not that it is true.
Decagon was founded in August 2023 and counts Rippling, Notion, Duolingo, ClassPass, Eventbrite and Vanta among its customers.
“Founded in August of 2023, their platform is already being used by large enterprises and fast-growing startups like Rippling, Notion, Duolingo, ClassPass, Eventbrite, Vanta, and more.”
Jesse Zhang started a company after graduating from Harvard, which was later acquired by Niantic, before he left to found Decagon.
“started a company right out of school. Uh, that company was eventually bought by Niantic, and then I left to start this company.”
Jesse Zhang grew up in Boulder, competed in math contests and studied computer science at Harvard.
“grew up in Boulder, did a lot of math contests, stuff like that growing up, studied C S at Harvard.”
Klarna's CEO said its AI assistant handled 2.3 million customer service chats in its first four weeks.
“his tweet basically said in the first four weeks, they handled 2.3 million customer service chats.”
Klarna's CEO reported the AI resolved customer issues in two minutes versus eleven minutes for a human agent.
“It resolved customer errands or issues in two minutes versus eleven minutes for a human agent.”
Klarna's AI assistant was reported to operate around the clock in 23 markets and 35 languages.
“instantly they were live twenty four seven in twenty three markets in thirty five languages because AI supports so many things.”
Elad Gil said Klarna shifted about 700 full-time agents to other work after adopting AI support.
“And I think, um, they sort of shifted 700 full-time agents to do other work, right?”
Klarna's CEO reported customer satisfaction on par with humans and a 25% reduction in repeat inquiries.
“The customer satisfaction was on par with humans. It's 25% reduction in repeat inquiries relative to people.”
Jesse Zhang says Decagon's customers first measure what fraction of conversations the AI agent can handle, then customer satisfaction and NPS.
“So for our customers, it's always the same. It's what fraction of total work, i in this case, like conversations, can the AI agent do?”
Zhang says Bilt Rewards, nearly a year into using Decagon, restructured its customer support team.
“And then now, uh, basically, we're almost almost a year in at this point. They've been able to really restructure their customer support team.”
Zhang says Bilt Rewards' published case study quantified savings of around 65 agent headcount.
“So far, uh, it's around sixty five agents of just like headcounts.”
Zhang describes the orchestration layer as combining multiple models, using evaluations that measure each model's strengths, around a customer's business logic.
“the orchestration layer is kind of how you can use all these different models together. You probably have evals set up that measure how good each model is at certain things.”
Zhang says Decagon uses language models to read every conversation, find knowledge gaps and categorize trends, since humans cannot review millions of conversations.
“If you have a million conversations, it's like, okay, like, no one's reading all those.”
Zhang says instruction following, rather than quantitative reasoning, is the type of model intelligence that matters most for Decagon's agents.
“the type of intelligence that matters the most, we would probably describe it as instruction following.”
Zhang says Decagon is experimenting with voice-to-voice models, with OpenAI doing significant work in the area.
“So nowadays you have the the voice to voice models that we're playing around with. Uh, OpenAI's doing a lot of work here.”
Zhang says voice-to-voice models have low latency but some production uses need extra computation to fetch data and make multiple model calls.
“Voice to voice latency is great. Uh, sometimes though, with these production use cases, you do need the extra computation cycles to, you know, fetch data, do multiple model calls.”
Elad Gil named Cartesia, ElevenLabs, OpenAI and Google as providers of text-to-speech and speech understanding services.
“there's companies like Cartesia, ElevenLabs, OpenAI, Google, et cetera, who are starting to provide some of these services and APIs.”
Elad Gil cited the teams at Braintrust, Pika, Cognition and Decagon as AI startups with founders from math and coding competition backgrounds.
“But I think the Brain Trust team and the Pika team and Cognition, which launched Evan, and then you all all kind of have that common thread.”
Zhang says he angel invests in several fellow contest-alumni startups, and their founders in turn invest in Decagon.
“I angel invest in a lot of the companies you just listed. Uh, a lot of their founders are angel investors in our company.”
Zhang says his co-founder Ashwin grew up in India and also took part in math and coding contests.
“my co-founder Ashwin also similar background. Um, he didn't grow up in the US. but in India, he did a lot of these contests as well.”
Zhang said Anthropic's computer use demo is, in his opinion, not yet ready for production use.
“we've seen the computer use demo from Anthropic. Uh, probably, in my opinion, not production ready yet.”
Zhang says buyers of customer support agents often test them on 1% of volume and scale up, benchmarking against alternatives.
“Let me just put you out into production for, you know, 1% of the volume and build up from there”
Zhang says incumbents such as Salesforce are also pursuing AI customer support agents and will be alternatives buyers benchmark against.
“a lot of the old school companies, like Salesforce, are gonna be this is a very exciting space for them too.”
Zhang argues that because models are non-deterministic, buyers will be slow to trust generative AI agents for security work that must catch every small event.
“because the models are in inherently non-deterministic, it's very hard for buyers to, like, really trust a gen AI solution there, and so especially, like, agentic solution.”
Zhang cites text-to-SQL companies as an example where agents seem promising but the return on investment is hard to measure because humans must still monitor them.
“there were a lot of, like, text to SQL companies, like, stuff like that, where you could kind of see it working.”
Zhang cites coding agents as an example of a use case that can be rolled out gradually without being perfect from the start.
“I think coding agents is, like, a good example of this, where, um, you know, you can just, like, section off some tasks for them, and, like, they'll do it.”
Zhang explains how talking to customers led him and co-founder Ashwin to customer service as the ideal agent use case, and why transparency defines the product.
“we kind of arrived at, you know, our current use case as maybe what we think is like the golden use case”
Compares Klarna's reported results with Decagon's own metrics and a case study showing a fast-growing company avoiding support hires.
“within, um, a basically a month of starting to use us, they were able to, you know, stop scaling their their team”
Explains that differentiation lies in orchestration and software for observability, not model access, and that instruction following matters most.
“the type of intelligence that matters the most, we would probably describe it as instruction following.”
Covers why voice is the next channel for support and the trade-offs between voice-to-voice models and text pipelines.
“you can always do something like, Hey, give me a sec, I'm looking up your data.”
Zhang says a tight community of contest alumni moved into startups, helping each other informally as investors and friends.
“startups have been a lot more mainstream, a lot of folks in this demographic have gravitated towards startups.”
Zhang expects agents that see the user's screen and humans who supervise and edit agents, with control tools as Decagon's edge.
“there's gonna be a lot more people that are ser supervising and editing agents.”
Zhang argues successful agent uses need gradual rollout and easily quantified return on investment, making him bearish on many near-term uses.
“you have to really be able to easily quantify the ROI.”
Hello and welcome to No Priors. Today I'm talking to Jesse Zhang, co-founder of Decagon. Decagon is an early-stage company building enterprise-grade generative AI for customer support. Founded in August of 2023, their platform is already being used by large enterprises and fast-growing startups like Rippling, Notion, Duolingo, ClassPass, Eventbrite, Vanta, and more. Jesse, welcome to No Priors.
Of course, thanks for having me. Elad.
Absolutely. Maybe we can start a little bit, um, with sort of your background and what Decagon does. You know, you're a serial founder, you started another company before this, Synthetic Bot. And, um, you know, now y- you and Ashwin have started Decagon and you've been working on it for a while and have seen some really interesting adoption from companies like Rippling, Notion, Eventbrite, Vanta, Substack, and many others, right? So you've you've really started to carve out a real, um, space for the company.
Can you tell us a little bit more about what Decagon does, how it works, what the focus is of the company?
Of course, yeah. So quick background on me. Um, grew up in Boulder, did a lot of math contests, stuff like that growing up, studied C S at Harvard. As you mentioned, started a company right out of school. Uh, that company was eventually bought by Niantic, and then I left to start this company. Uh, Ashwin and I, we we met through mutual friends, um, officially met at this, uh, V C offsite. And when we got together, we were like, okay, um, biggest learning from first company is that you can't really overthink things too much.
We, um, we started by just kind of obviously being interested in AI agents. It's very exciting technology, arguably like the coolest thing from from this generation. And, uh, we just, you know, talked to a bunch of customers, like the ones you listed. Um, we, I think, over the years, have gotten a lot better at figuring out, you know, how to talk to folks and, you know, how, what, what questions to ask. And, uh, through that process, we kind of arrived at, you know, our current use case as maybe what we think is like the golden use case for for these AI agents, which is customer interactions, customer service.
Um, the the use case is very tailor-made for for what LLMs are good at. And so we started building from there, right? And we still w- weren't thinking too much about, you know, the vision or anything yet. It was just like, alright, we had a lot of customers in front of us. How can we make it so that they're happy and, uh, they really like what we're building? And then that led to kind of where we're at now.
I would say right now, as a company, Decagon, we ship these AI agents for folks to use on the customer service, customer experience side. The thing that's made us special so far is, um, we have a huge sort of focus on transparency, I guess. So when people use us, especially these larger companies, it's it's very important for them that the AI agent is not a black box. That they feel like, okay, even though LLMs are cool and, like, you know, there's there's a lot of things you can do with them, that they can see how decisions are being made, like what data is being used, how do you come up with answers?
And if I wanted to get feedback, I can, that sort of thing. So currently, we're in production with a bunch of these these large folks that have large support teams. Um, pretty much any company that has a large, sizable support operation is is a good fit for us.
That makes a lot of sense. It's interesting because, um, I feel like one of the things that w- that's been really striking over, say, the last year in the AI world is the CEO of Klarna, um, uh, posted on X, or tweeted about the impact that AI has had on its on their customer support or service team. And Klarna is sort of like a buy now, pay later service out of Europe. And, um, you know, his tweet basically said in the first four weeks, they handled 2.3 million customer service chats.
The customer satisfaction was on par with humans. It's 25% reduction in repeat inquiries relative to people. It resolved customer errands or issues in two minutes versus eleven minutes for a human agent. Um, and instantly they were live twenty four seven in twenty three markets in thirty five languages because AI supports so many things. And so, you know, had huge impact on that company. And I think, um, they sort of shifted 700 full-time agents to do other work, right?
Um, in terms of the impact at Klarna itself as an organization, what sort of impact have you been seeing with your customers as they adopt this sort of technology? And how do you think through the lens of of, you know, what you're really bringing to these customers and the sort of satisfaction that their own end users have?
Hey, it's an interesting way to think about it, which is, um, you know, all these people are shipping this use case, right? It's like, you know, there's there's a lot of evangelists out there, which is nice. Uh, I mean, the Klarna article is is awesome. It gave there's a lot of tailwinds for the for the industry. Um, and I think one interesting thing we've seen is that the benefits that people get are are all roughly in the same vein, but different people prioritize different things.
And so at this point, it's not really even that much of a hot take to say that, like, in a couple years, these agents are gonna be super pervasive. Uh, people are gonna use them for all these customer interactions. They're gonna be everywhere. And so, to your point, like, what what is the benefit? So for our customers, it's always the same. It's what fraction of total work, i in this case, like conversations, can the AI agent do?
So how how much work is this saving us? And then two, um, how much happier are our customers, right? Like, what's the customer satisfaction score? Um, NPS score. Those two are often just the leaders by far. As I said before, like, different people maybe val- value each one slightly differently. And then there's kind of other things like, okay, well, we wanna make sure that there's accuracy, right? Like, if we're in a regulated industry, uh, you know, this this has to be very accurate for us.
And so that those are kind of where the benefits lie. It's like we're saving a bunch of money, we're saving kind of time and resources. But also on the other side, we're kind of making the customers happier, and so that can lead to higher retention, uh, more conversions, and it's kind of a lot more upside there. It's like you're giving every customer a personal concierge, basically, in their pocket that they can chat with any time, any language, twenty four seven, and, you know, that's that's that can be pretty transformational for a lot of businesses.
Is there any, um, example customer that you can talk about as a case study in terms of the impact this has had, how it's lifted their metrics, the success they've seen using Decagon?
Of course, yeah. So we we just did a big case study with a company called Built Rewards. Uh, great use case for us. They have a very large user base, growing very quickly. You're actually using it to either make points or make payments. Uh, a lot of my friends use the product. And th- then as a result, because you have a large customer base, like people have questions, people will have things that they need help on.
And so, like, the the the number of support inquiries basically grows linearly with the, uh, number of number of users. And because of that, because they're growing so fast, and basically exponentially, that means the number of support queries is also growing exponentially. So when they first started using us, that was like the main goal. It's like, holy crap, we're getting overwhelmed by all this volume. Can AI help here? Uh, and so the thing that ended up happening there was, yeah, within, um, a basically a month of starting to use us, they were able to, you know, stop scaling their their team, and, um, the AI would take over a lot of the automation, just makes everything very smooth.
And then now, uh, basically, we're almost almost a year in at this point. They've been able to really restructure their customer support team. And again, we we published a case study on this where they were able to quantify, like, okay, what are the savings, right? And so so far, uh, it's around sixty five agents of just like headcounts. Say, so very tangible, uh, difference. And then for us, it's also great because we're able to, um, you know, provide them the value there.
It's like a very easy ROI. But the customer experience is also a lot snappier. And, you know, they get a lot of social media posts about like, holy crap, like, I just tried, you know, Built Rewards support thing, and it's doesn't feel like any sort of AI or chatbot system we've ever used before. So, um, that makes us happy.
Could you tell me a little bit more about you built from a technology and infrastructure perspective? So I guess there's the core models that anybody can access, right? The the GPT four O's of the world or GPT four, or the Claude Sonnets, et cetera. Um, and then there's all the stuff you've built on top of it to actually make this work well for your specific use case and for, um, customer support agents. Could you could you tell us a bit more about what you all have had to build over time?
Of course. Uh, like you said, everyone has the same access to the same models. We see ourselves very much as a software company. And we're obviously doing a lot of work around AI and, like, using the AI models a lot. But I would argue that most, you know, applications nowadays are they're real software companies, and AI models are kind of tools that everyone can use. And so most of the sort of alpha, or most of the specials stuff that you build is on top of models.
It's either the orchestration layer or the software around it. Uh, for us, there's been a big focus on both. Uh, the orchestration layer is kind of how you can use all these different models together. You probably have evals set up that measure how good each model is at certain things. You put them together, and the whole goal of putting them together is to mold it around the business logic of the customer. That's part one. The other thing you build is just very classic software, right?
You have this AI agent there. It's all the things I was saying before, like, you know, transparency is a big piece. You it's you really don't want this to feel like a black box. That's just they're answering questions. And so, how can you build all the tooling to see, like, okay, what's the data that the agent's using? How's what steps is it taking? Um, can I analyze all these conversations that are coming in? If you have a million conversations, it's like, okay, like, no one's reading all those.
So, how can you make it so that the A AI, the LLM, can read every single conversation, tell you how stuff's going, find gaps in their knowledge, um, give you a breakdown of like, okay, here are the big categories you should care about. There's, like, a bit a trend here that that's been interesting. So that's all the software around it that that we're building. Um, and that's typically how it's structured. And and the orchestration layer, I think is gonna be different for every agent, right?
Like, our agent versus, like, a coding agent, that orchestration's gonna look pretty different. Uh, but at the end of the day, you're kind of just building a sort of a a structure on top of the LLMs.
Yeah, it seems like we're very early in the days of, um, true agentic stuff. And that includes the ability to, um, sequence chains of events that includes certain forms of reasoning. Obviously, there's things like, um, o one and other things that have been coming out to to start to try and address this, but we seem quite early in the scaling curves. What do you think are the main pieces of technology that are missing to really take you or your sort of vision to the next level in terms of how these agentic systems should work?
Yeah, so one thing we were talking about the other day is there's actually different types of intelligence, uh, with the AI models. And a lot of the recent developments with o one or Sonnet and and stuff like that has has been around, I guess, like, quantitative reasoning intelligence. So they've gotten better at coding. They've gotten better at math. Um, and for us, actually, those things help, but they're actually not the biggest difference maker. So, in our use case, the type of intelligence that matters the most, we would probably describe it as instruction following.
So you just have a bunch of instructions, like, can you follow it to a t? And I'm I'm sure there's other kind of types as well, but, um, for us, we're excited to see developments on the other areas too. Um, and people just everyone's saying, like, oh, there's a plateau happening with, like, you know, the core models and the intelligence. I think when when most people say intelligence like that, they're probably talking about the the reasoning capabilities.
Um, for us, in, like, the agentic flows that we use, instruction following is a huge piece. Because you have to, like, you know, just think about, like, a customer service, you know, SOP, or like a play, like a workflow or something like that. Like, you just have to be very accurate about it. Um, and so I know there's research going on about this in the in the major labs, and I think that's one thing we're looking forward to next year.
One one other area that, um, it seems like really, um, touches on customer success and customer support and sort of user experience is also voice-based support. And I think one of the things that's a little bit under-discussed in the AI world, cuz we keep talking about large language models and understanding of text and all, and obviously that stuff is crucial to everything else, but I feel like we almost under-discuss, uh, text-to-speech engines and the ability to understand spoken word and then respond with audio, right?
And so there's companies like Cartesia, ElevenLabs, OpenAI, Google, et cetera, who are starting to provide some of these services and APIs. How much of an impact has that happened? What you're doing, or is that a separate type of product, or, you know, how do you think about the voice component of these things?
Great question. A huge impact. So we, um, have customers now trying our voice agents. And if you just think about our space, right, like, you have the the overall problem is the same, which is you have a bunch of customers, they have, you know, questions or issues or things you need to talk about. And the channel really doesn't matter for them. It's like some people prefer voice, some people prefer chat, some people prefer email, some people prefer SMS or something like that.
And so our job is to handle all of those. Um, and obviously you you start with text because that's the most it's like the easier one. It's like easy to evaluate for the customer as well. I think just now you're getting to the point where you have big companies that are very interested in voice. And like, actually, they they've seen the results of a text-based agent, and they're like, okay, well, yeah, you should we should be able to generate voices and do the same thing for phone calls.
Um, uh, none of this would be possible without the the models that you just listed, right? And the the those companies, so, like, LLM Labs, um, OpenAI is doing some cool stuff, Cartesia. Uh, and I think there's also been huge strides this year with those models around, like, how real realistic the voices sound. Also, like, latency matters a lot in our use case, cause if you're making a phone call, like, you expect things to feel very snappy.
Um, so, yeah, big, big, uh, big topic for us. And as these companies get better, I mean, we're working with them pretty closely right now on how how you can actually, like, build these things well at scale. Uh, but as they get better, that's also gonna be huge for us to keep delivering these voice agents.
Makes sense. Yeah, my sense is one of the issues is latency in terms of, um, it takes, uh, enough time to take a audio stream where somebody's talking, translate that into text, feed that into a language model, and then output it as, uh, voice again. That it feels, um, there's a lot of pauses, or people have to kind of wait. And there's different things that people have been trying to do in the background, like streaming the potential solutions back out and then, you know, being able to try and shorten that latency timeline.
Do you feel do you feel latency is still an issue, or is it just solved by integrating voice directly into the models in a deeper way for for some of these services? Or w- when do you think latency becomes a solved problem for these sorts of application areas?
I mean, lat- latency is a big, big deal here, of course, uh, with voice models. So nowadays you have the the voice to voice models that we're playing around with. Uh, OpenAI's doing a lot of work here. I think there's obviously a lot of trade-offs there. Voice to voice latency is great. Uh, sometimes though, with these production use cases, you do need the extra computation cycles to, you know, fetch data, do multiple model calls. Um, or y- there's there might be other reasons that you you can't do voice to voice.
And so okay, that's that's one option that you you would consider. Uh, the other one is the the one you described where you're kind of going through, you're transcribing, or, yeah, doing speech to text, and then doing all the computation within text, and then generating the voice at the end. That always causes a little bit of extra latency, of course. And so as you mentioned, a lot of folks have figured out fairly clever ways to to get around that.
You can, you know, start generating stuff first. Um, in in our use case, you can always do something like, Hey, give me a sec, I'm looking up your data. Um, so the these are all things we're playing around with. And I think for each customer that we work with, there's different trade-offs. And so we're really trying to base what we build, uh, on the things that we're hearing from them and, you know, the the sort of priorities that they have.
That's cool. One thing that I think is kind of interesting is the number of companies in the AI world today that have been founded by people with mat math Olympiad or IOI or or other sort of backgrounds, right? And I think you were, um, sort of, uh, involved with math Olympiad stuff in high school. Uh, I think Decagon has actually hosted some math Olympiad events for the team, um, which isn't like your typical happy hour. But there's other other teams and companies.
I mean, before that there was ramp and things like that. But I think the Brain Trust team and the Pika team and Cognition, which launched Evan, and then you all all kind of have that common thread. Where do where do you think that comes from? Like, why do you think this community is now so active in AI?
Uh, it's a good question. I mean, yeah, we're we're actually all around the same age as well, so we've known each other since, like, you know, middle school, high school. Um, well, one it's it's a great community. Uh, for us, you know, we have a lot of people on the team with math contest, coding contest backgrounds. I think it's more so that this community was always there. Like, math contests have been around for a while, and a lot of super smart kids that go through that.
Uh, it's also a great way for folks to kind of, you know, get to know each other and get connected and, you know, build friendships. And I think the main thing is that now, in the last few years, maybe the last, like, you know, five, six years, like, there's just because startups have been a lot more mainstream, a lot of folks in this demographic have gravitated towards startups. As opposed to, you know, traditionally, it'd be, you know, either academia or quant trading and and things like that.
So there's just a big influx of these super smart, super talented people that come into the startup world. And because there's this community aspect, uh, you know, folks can see what other people are doing and, like, you know, what what sort of works and types of companies that people are building, um, that uh, uh, no, I wish it didn't say they're all the same, but I think a lot of folks with these backgrounds are now kind of working on startups.
And that's why there's a lot of, you know, I guess, pro progress in the companies that folks folks have been building.
Uh, and are there are there ways that you all have been sort of supporting each other through the startup journey? Cause I feel like every generation, there's sort of a clique of people who built some of the more interesting companies who all kind of interact. They provide advice, maybe they angel invest in each other, like, there's kind of a thriving community. And every five to seven years, it kind of shifts who it is. And I feel like, you know, the IOI sort of math Olympiad community or coding competition communities are kind of very engaged right now.
How is there any formal version of that, or are you all just kind of informally helping each other?
Uh, yeah, I mean, I angel invest in a lot of the companies you just listed. Uh, a lot of their founders are angel investors in our company. It's it's very informal. Obviously, it's just like, you know, casual friends helping each other. I I think the the main thing is that, uh, with company building, there's just a lot of lot of surface area, right? As you know, there's just like, how do you hire people? How do you, like, do sales?
Like, how do you build this thing? Like, how do you structure, like, comp? Like, I don't know. Uh, there's infinite things. So, yeah, having the other data points is obviously super helpful. So I hang out with them quite often, play games, play card games. It's a Chinese version of bridge that I play with a lot of these folks quite often. And, uh, it's just it's fun. We just kinda hang out. Everyone's kind of in the relative same stage of life.
And so, yeah, like you said, there is definitely a lot of camaraderie and and help that that goes around.
Is coming from this background from the sort of math Olympiad community impacted at all how you think about hiring or your hiring practices at Decagon?
Uh, a little. I mean, if if someone else has the same background and, you know, has gone through the same either contests or programs, obviously that is pretty good signal since I have a good idea of, like, what those those people have done. Uh, my co-founder Ashwin also similar background. Um, he didn't grow up in the US. but in India, he did a lot of these contests as well. And so, um, yeah, there I think there's some correlation with people who just as kids just did a lot of this stuff.
And then, you know, now we're all adults. And, you know, there's there's some sort of, you know, signal there when you're when you're talking about hiring. But, uh, for the most part, like, it's there's there's so many talented people here, whether or not you, you know, did math contests or not. In SF, you know, at DecafCon and at other companies that I think our hiring process has been more or less the same. It is like a nice, um, sort of trigger for events, I guess.
So when you host these events, you know, people come out and you can get a nice community of, uh, folks that are interested in the same things. And, uh, we're gonna be hosting probably more, uh, and some not not all them are gonna be, like, contest based. Obviously, like, you know, puzzles and things like that, where, you know, you just get a lot of fun engineers and, you know, people bringing their friends, and that's that's pretty important to us.
And then I guess, um, for AI writ large, what do you what are you most excited about in the coming years? Or if you were to extrapolate out twelve to twenty four months, what are you anticipating most keenly, or or what are you waiting for?
Uh, so obviously the model's getting better is awesome. Uh, the models getting better across different modalities, also awesome. We talked about voice. There's also other parts, the other modalities that also tangentially interesting to us, right? So you talk about a lot of our companies are, um, a lot of our customers, you know, have software products. And so it'd be awesome if, you know, you're asking questions to the AI agents, and it has, like, context of your entire screen and, like, all the interactions you've done and stuff like that.
Like, that would be great. And you can even go a step further and have it actually, like, help you navigate stuff. So there's just so much you can do there where you talk about the other modalities or even just more advanced model capabilities. Like, you we've seen the computer use demo from Anthropic. Uh, probably, in my opinion, not production ready yet. But as that gets better, there's a lot of, you know, cool things you can do there.
So on the model side, that's one thing we're excited about. Uh, on the it's not, like, core model side, I think one thesis we have is as, you know, the years, you know, go by again, AI agents, I think at this point, undeniable that there's gonna be a a reasonable expl- explosion of them where there's used on a bunch of different use cases. I think some use cases will take longer than others, but the value that they're providing is is pretty undeniable.
So there's definitely gonna be a lot of AI agents out in the world, you know, in our use case, customer service, and other use cases. Uh, but one th- one thesis we have is that the nature of the work of, you know, human agents and, you know, people like us, is also gonna change pretty drastically. And one of the things that are gonna change is that there's gonna be a lot more people that are ser supervising and editing agents.
And so that's something we think about. Um, we're excited for a lot of the sort of innovations there because, uh, right now, a big part, like I said before, we care about letting, you know, the human agents, on for our customers and their leadership team to go in and make changes and, like, monitor the agents. Just have a lot of, like, visibility and control. And what does that look like, right? Like, if you compare it to human, if you're monitoring a human, you kind of you can give them feedback in real time.
You can be like, Oh, no, no, don't, don't, don't do this. Like, you did this thing wrong. Like, please do this next time. When you're doing that with the AI agent, there's, like, a lot of different possibilities. Because, you know, they have some things that are different than humans, right? They're infinitely scalable. They're, um, they're like, you can you can, like, really hard code things sometimes. So, um, that's the other area probably going to next year that we're we're looking forward to.
That's really cool. And, um, do you view that as the main area of differentiation for you relative to some of the other folks in the market providing customer success and support tooling?
Yeah, right now, that's that's probably the the biggest thing. Uh, when we the interesting thing about our space is that, and I think this will probably be true for a lot of AI agent spaces, is that the results are very quantifiable. Like, you're you're basically taking the agent, and you can you're benchmarking against, okay, how good would a human be? And, like, how much money is this saving me? Like, how much better quality is the customer experience?
And so because of that, uh, when people evaluate us in our space, it's it's pretty, like, quantitative evaluation. You're like, Okay, cool, this kind of works. Like, let me just put you out into production for, you know, 1% of the volume and build up from there, and maybe do that with a another option. Or, you know, a lot of the old school companies, like Salesforce, are gonna be this is a very exciting space for them too.
So they're gonna have alternatives. And then you just benchmark everyone, right? Like, how good are the stats? How good are the metrics? Like, how good of a job is everyone doing? And, uh, I think so far we've been performing very well. And the the main reason for that is this sort of transparency piece, you know, giving people observability, explainability, control over the AI. And there's still a long way to go in that field, right? Like, there's still so much more you could do.
Uh, and that's that's been our specialty so far.
That's great. Yeah, no, I've I've had some conversations with your customers over time as people have been trying some of these agents. They've called me to ask questions about different companies in the space and everything else. And the three things they tend to point out is you you all ship really fast. Uh, you're very responsive as a team and company, not the third, and most importantly, just the product tends to outperform. And so I think that's really been, um, you know, uh, great to watch over time.
How do you think about the areas where AI agents are gonna be, um, successful versus not successful in the short run?
So basically, o- one of the things that we have been thinking through, and this is something that was pretty big for us when we were first starting out, is that a lot of there's gonna be a huge variance between, like, the different types of AI agents and how successful they'll be, and, like, how quickly they'll take the rollout. Because when we were first starting the company, right, like, uh, we we were pretty open to what to build, and we knew that AI agents was exciting.
Uh, at that point, we didn't even know that if there would be any, like, real use cases that would emerge even in the next twelve or twenty four months. But we're kind of exploring. I think our view is that for the vast majority of use cases right now, it is still, like, like, there's not gonna be real commercial adoption, uh, with the state of the current models because of a bunch of things. So one one big thing is that if in a lot of spaces, you can't, there's really no, like, structure there to, like, incrementally build up.
Like, it has to be good, like, almost perfect off the bat. So if you think about, like, a space, like, you know, security or or or something like that, where, okay, okay, you have all these, like, sims out there, and it's like, it makes sense. There's, like, tons of logs. Like, that's that's perfect for AI models. Uh, but the goal of that job is, like, you need to catch, like, any small thing that happens. And so because the models are in inherently non-deterministic, it's very hard for buyers to, like, really trust a gen AI solution there, and so especially, like, agentic solution.
So, like, I think that adoption there is gonna be really, really, really slow, a lot slower than people think. Even though, like, people have cool demos and, you know, see things seem to work, like, just getting really enterprise adoption is gonna be very slow. Um, so that's that's like one interesting thing we've been thinking about. And the other side of that is that, uh, there are also a lot of spaces where, you know, on the surface, it seems like, oh, like, AI agents will be perfect here.
But then, uh, the sort of follow-up is that it's actually not that easy to quantify the ROI that's happening. Um, I would, you know, one example I would give this is like, uh, you know, there's there were a lot of, like, text to SQL companies, like, stuff like that, where you could kind of see it working. But basically, immediately, everyone's reaction is, Oh, this is cool. But, you know, we're still gonna have to have someone, like, monitoring it and, you know, editing it.
And so it becomes kind of a co-pilot. Okay, cool. So then, like, how do we measure, like, how much we should pay for one of these agents? Um, it's very difficult because, like, most teams don't have that many data scientists anyways. And so if you're if you're claiming that you have a AI agent data scientist, it's like, okay, let's benchmark you against a real one. You're probably not gonna be able to replace a real one. So I think that's the sort of thing where it's like, it's very hard to quantify the ROI.
Like, you're saving some people time, but because of that, like, it's it's, you know, like, a if you have if I'm a large company, it's hard for me to justify, okay, I'm gonna pay you a large contract for this, like, AI agent data scientist. Um, so I think those are the things that we were thinking through. Uh, or not thinking through, like, in in the moment, we're obviously just, you know, asking customers, like, what what their willingness to, you know, to invest in certain things is.
But in hindsight, I think we're looking back on the last year, that's been a big thing that's been true, which is the n- the use cases that emerge. Like, you have to have those two qualities, like, it it has to be able to be something that can be rolled out slowly and doesn't have to be perfect off the bat, but it's already providing value, right? Like, I think coding agents is, like, a good example of this, where, um, you know, you can just, like, section off some tasks for them, and, like, they'll do it.
Uh, and the other piece is, uh, the ROI. Like, you have to really be able to easily quantify the ROI. And in our case, luckily, you have these support agent teams and, um, you know, people track metrics very closely. So that that's something we've been thinking about. I think most I think the takeaway from that is, like, yeah, probably more bearish on a lot of these AI agent use cases in the near term. But I think as these models get better, there's gonna be they'll unlock a lot of new cases.
Super interesting. Jesse, thank you so much for joining us today.
Thanks a lot. Thanks for hosting. It's great seeing you.
Find us on Twitter at No Priors Pod. Subscribe to our YouTube channel if you wanna see our faces. Follow the show on Apple Podcasts, Spotify, or wherever you listen. That way you get a new episode every week. And sign up for emails or find transcripts for every episode at no dash priors dot com.