Jan 16, 2025 · 30m · no-priors
No Priors Ep. 97 | With Decagon CEO and Co-Founder Jesse Zhang
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Decagon CEO Jesse Zhang joins Elad Gil on No Priors to discuss how enterprise AI agents are revolutionizing customer experience through specialized orchestration architectures, measurable operational ROI, and human-in-the-loop supervision.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 23% of the talking time here. How this is scored →
The hosts as informed peer 4.8 Guest teaching 3.8 Guest disagreement 0.3 The hosts pushing back 0.2
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 10:46 Contrasting instruction following against pure reasoning
Jesse politely pushes back against the broad hype surrounding frontier reasoning models like o1, explaining that customer support agents require strict instruction adherence rather than math or coding reasoning.
Hardest push from the hosts ▶ 14:29 Elad questions voice latency resolution timelinesElad presses on whether pipeline latency remains a fundamental hurdle in conversational voice AI, requiring deeper integrated models rather than chained APIs.
Biggest teaching moment ▶ 25:33 Breakdown of non-deterministic risks in security and data agentsJesse delivers an incisive analysis of why enterprise buyers reject text-to-SQL and cybersecurity agents due to the inability to roll them out incrementally or prove clear replacement ROI.
The host holds their own ▶ 3:15 Elad details Klarna's AI agent operational metricsElad establishes strong host authority by reciting exact data points regarding chat volume, speed improvements, and 700 redirected headcount from Klarna's rollout.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Evaluating the Organizational Impact of Customer Service AI | 6 | 2 | 0 | 0 | Elad demonstrates domain knowledge by citing specific operational metrics from Klarna's AI rollout. Jesse agrees with the premise and details how enterprise clients prioritize resolution volume and CSAT scores. | |
| Case Study: Scaling Support at Bilt Rewards | 4 | 4 | 0 | 0 | Elad prompts Jesse on customer ROI and architectural layers above foundational models. Jesse explains Decagon's orchestration layer, automated conversation analytics, and case metrics with Bilt Rewards. | |
| Instruction Following Versus Reasoning Intelligence | 6 | 5 | 1 | 1 | Elad brings up reasoning models and technical voice latency bottlenecks across speech-to-text pipelines. Jesse offers a nuanced distinction between reasoning benchmarks and enterprise instruction-following accuracy. | |
| The Math Olympiad Community in AI Startups | 5 | 2 | 0 | 0 | Elad maps out the concentration of Math Olympiad alumni founding major AI companies. Jesse confirms the social cohesion of this peer network and discusses how contest problem-solving translates into early startup hiring. | |
| Future Frontiers: Multimodal Context and Human Supervisors | 4 | 4 | 0 | 0 | Elad explores upcoming AI frontiers and product differentiation. Jesse outlines a transition toward human supervisory roles and multimodal UI context where agents act on real-time screen interactions. | |
| Analyzing Commercial Viability Across AI Agent Domains | 4 | 6 | 1 | 0 | Jesse provides an analytical breakdown of why certain agent domains like cybersecurity SIEMs and text-to-SQL data science tools struggle with commercial adoption due to non-deterministic risks and unquantifiable ROI. |
Statements from this episode (11)
Zhang: Customer service is the golden use case for AI agents
“Our current use case as maybe what we think is like the golden use case for these AI agents, which is customer interactions, customer service. The use case is very tailor made for what LLMs are good at.”
Decagon AI saved Bilt Rewards 65 support agent headcounts in one year
“Now 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 published a case study on this where they were able to quantify like, okay, what are the savings? Right. And so, so f…”
Zhang: Most AI application alpha lies in orchestration and surrounding software
“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.”
Zhang: Instruction following matters more than reasoning for customer service AI
“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.”
Zhang: End-to-end voice AI falls short when complex tool use is needed
“Voice-to-voice, latency is great. Sometimes, though, with these production use cases, you do need the extra computation cycles. So, you know, fetch data, do multiple model calls or there's, there might be other reasons that you can't do voice-to-voice.”
Zhang: Math Olympiad talent has shifted from quant trading to startups
“In the last few years, maybe it lasts 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 eithe…”
Zhang reveals reciprocal angel investing among Math Olympiad AI startup founders
“I angel invested in a lot of the companies you just listed. A lot of their founders are angel investors in our company.”
Zhang: Anthropic's computer use capability is not yet production-ready
“Like we've seen the computer use demo from Anthropic. Probably in my opinion, not production ready yet”
Zhang: Most AI agent use cases lack commercial viability with current models
“I think our view is that for the vast majority of use cases right now, it is still, like there's not going to be real commercial adoption with the state of the current models because of a bunch of things.”
Zhang: Enterprise adoption of cybersecurity AI agents will be slower than expected
“And so because the models are inherently non-deterministic, It's very hard for buyers like really trust a gen AI solution there. And so, especially agentic solution. So like, I think that option there is going to be really, really, really slow, a lot slower th…”
Zhang: Viable AI agent use cases require incremental value and quantifiable ROI
“The use cases that emerged, like you have to have those two qualities. Like it has to be able to be something that can be rolled out slowly and doesn't have the perfect off the bat, but it's already providing value, right? Like I think coding agents is like a …”