Sep 19, 2025 · 59m · 20vc
20VC: Why 90% of Founders Build Startups Wrong | Why AI Growth Rates are Sustainable & Remote Work is BS and the AI Talent War | Competing with Brett Taylor and Sierra: Who Wins the Customer Service War with Jesse Zhang, Decagon
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of The 20VC Podcast, host Harry Stebbings interviews Jesse Zhang, co-founder and CEO of Decagon, exploring how Decagon rapidly scaled to a $1.5 billion valuation by automating enterprise customer experience. Zhang shares critical lessons on enterprise sales, unlocking labor budgets, building high-density in-person startup cultures, and navigating modern AI valuations.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 34.3% of the talking time here. How this is scored →
Harry as informed peer 5.1 Guest teaching 5.8 Guest disagreement 1.9 Harry pushing back 2.9
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 45:15 Rejection of corporate wellness and stress mitigation culture
Jesse forcefully rejects the conventional narrative around stress management, arguing that corporate wellness efforts do more harm than good and that founders should embrace stress directly.
Hardest push from Harry ▶ 23:45 Challenging the 60-80% ticket resolution spreadHarry explicitly refuses to accept Jesse's broad resolution metric, calling the gap between 60% and 80% a very large chasm and demanding a clearer long-term trajectory.
Biggest teaching moment ▶ 36:03 The operational dangers of accepting inflated valuationsJesse delivers a thorough masterclass on startup mechanics, detailing how accepting peak valuations dilutes employee achievements, complicates equity recruitment, and risks turning functional businesses into zombie companies.
Harry holds his own ▶ 20:29 Contrasting PLG commoditization with labor budget captureHarry demonstrates deep SaaS domain knowledge by contrasting $20 PLG code editors like Cursor with enterprise top-down labor budget replacement to test Decagon's pricing power.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Math Olympiads and the Founder Mindset | 3 | 5 | 1 | 1 | Harry asks about the connection between Math Olympiad participation and founder success, proposing a mock $1M niche venture fund. Jesse explains how reasoning capability and problem-solving skills translate directly into startup execution. | |
| Lessons from Early Startup Success and Failures | 4 | 6 | 2 | 3 | Harry quotes Jesse's past talk regarding market selection and presses him on whether execution trumps market selection. Jesse reframes the question using a piano-learning analogy to explain how discovery execution helps founders identify real willingness to pay. | |
| Vision Selling vs. ROI Demonstration in AI | 6 | 5 | 3 | 4 | Harry quotes Scale VC investor Rory O'Driscoll regarding Harvey selling vision ahead of product and asks if AI SDRs oversold. Jesse distinguishes law firm market dynamics from enterprise customer experience, while Harry pushes back on signaling risk and talent investor optics. | |
| Building AI-Native Platforms vs. Legacy SaaS Architecture | 5 | 6 | 1 | 2 | Harry asks a structured framework question about the road to feature parity versus a complete paradigm shift. Jesse explains why legacy SaaS tools like Zendesk carry architectural baggage compared to AI-native Agent Operating Procedures. | |
| Transitioning from Software Budgets to Labor Budgets | 6 | 6 | 2 | 3 | Harry contrasts Cursor's $20 PLG commoditization pressure with enterprise top-down labor budget replacement. Jesse details how application-layer software captures higher multiples by benchmarking against labor savings rather than software cost-plus margins. | |
| Ticket Resolution Rates and the Proactive AI Concierge | 5 | 6 | 2 | 4 | Harry pushes back on Jesse's 60-80% ticket resolution range, calling it a very large chasm and asking where it lands in 3-5 years. Jesse outlines the transition from reactive ticket resolution to proactive brand concierge systems. | |
| Enterprise Guardrails, Customization, and In-House Building | 6 | 6 | 2 | 4 | Harry references Robinhood CEO Vlad Tenev building customer support in-house to challenge whether large enterprise clients scale out of vendor products. Jesse explains why non-fintech enterprises lack the developer tooling and observability layers required to maintain internal AI systems. | |
| AI Code Generation, Talent Wars, and In-Person Culture | 5 | 5 | 2 | 3 | Harry cites code generation metrics from Benioff and Vlad, then asks how Decagon competes for talent against mega-funded AI labs like Anthropic. Jesse explains their target candidate profile and why in-person work intensity acts as a talent filter. | |
| Competitive Dynamics and the Dangers of High Valuations | 6 | 7 | 2 | 4 | Harry asks why Salesforce's Agentforce hasn't succeeded as expected and quotes Jesse's prior interview regarding valuation traps. Jesse details turning down 1.5x-2x higher valuation offers to avoid demotivating employees and creating a zombie company. | |
| Enterprise Market Makeup and Systems of Intelligence | 5 | 6 | 2 | 3 | Harry compares Salesforce and HubSpot market distributions to question if CX AI is winner-take-all. Jesse reframes AI software as a system of intelligence that stores business logic rather than traditional system-of-record configuration data. | |
| Hiring for Clock Speed, Culture, and Embracing Stress | 6 | 6 | 2 | 3 | Jesse shares contrarian views on hiring for pure clock speed over domain experience and argues that corporate wellness trends make stress worse. Harry agrees passionately, citing Revolut founder Nick Storonsky's view on winning team cultures. | |
| Founder Self-Reflection and Go-To-Market Innovations | 4 | 5 | 1 | 1 | Harry asks Jesse to reflect on his operational weaknesses and preferred investors. Jesse admits to struggling with micromanaging deal-level details and highlights Avra investor Anu Hariharan. | |
| Quick-Fire Questions: AI Hype, Model Access, and Leadership | 5 | 6 | 2 | 2 | During rapid-fire questions, Harry asks whether Jesse would choose exclusive access to leading AI models or unlimited top engineering talent. Jesse emphatically picks top engineering talent, explaining that application-layer value comes from system orchestration around models. |
Statements from this episode (39)
Decagon raised over $230 million at a $1.5 billion valuation
“Now, as one of the fastest growing companies in the valley, they've raised over a two hundred and thirty million dollars, with the last round pricing them at 1.5 Billion dollars.”
Zhang: Combining math Olympiad reasoning with sales skills is underrated
“If you can really unlock that level of, you know, reasoning capability and just like smartness and combine it with folks that can, you know, teach them how to sell or how to build a company, I think that's a pretty good combination. So that's something I think…”
Zhang: First-time founders fail by over-intellectualizing market narratives
“I think it's just a very common pattern where you start and you're just like super excited. You're listening to podcasts like yours, or just reading all these articles and you just, it's like really easy to form these narratives in your mind. When in reality, …”
Zhang: Prior exits empower second-time founders to target larger ideas
“I think the second time around, you're a little bit more grounded and you're also swinging a bit larger because you feel like you have a win under your belt. And so if you have another one of those, it doesn't really matter. And so you're just gunning for bigg…”
Zhang: Founders quit prematurely because discernment improves faster than execution skill
“Generally when you try new things, where you learn new things, the rate at which you're able to discern good versus bad improves a lot faster than your actual skill level. A classic example of this is like if you're learning piano or something, right? Like the…”
Zhang: Early startup execution serves primarily to discover viable markets
“In the early days, execution should help you find the right markets, because if you're actually executing your discovery well, you should be able to discover, okay, which markets actually will be real versus not.”
Decagon found enterprise CX buyers willing to pay six figures upfront
“When we started talking about, you know, the CX space and actually deploying conversational AI agents to actually talk with customers, people were able to justify that way more, right? They're like, oh, wow. Yeah, this would be useful because I have, you know,…”
Zhang: Enterprise AI buyers demand fast ROI proofs over vision sales
“You can't just go in and sell the vision because no one needs to buy into the vision. I think everyone already believes the vision. And so then it's more about showing the results quickly and being able to demonstrate that, hey, there is ROI here and that, you…”
Zhang: Decagon and Sierra will out-execute Salesforce on AI products
“I would expect a Decagon and a Sierra to be able to execute faster on a product side than Salesforce, but Salesforce has like so much distribution.”
Decagon raised seed funding from Andreessen Horowitz before having an idea
“The seed round we raised before we had any ideas. So we raised from Andreessen Horowitz.”
Zhang: Top engineering candidates care heavily about venture capital investor brands
“I mean, I just think that most people care. If you're responsible about your career, like, you're gonna look at all the different factors and Maybe the sort of kernel of truth behind that quote is that that shouldn't be the only thing they care about. But if y…”
Zhang: Venture capital platform teams cannot accelerate product-market fit
“I think it's basically impossible for any VC to help accelerate the process of getting to PMF. I think the issue is that VCs will sell that they can, but that's not possible.”
Zhang: Founders should pick different lead investors for Seed and Series A
“I actually think given a choice, you should have different investors for the C and the A because you just get two firms instead of one. They bring different things.”
Zhang: Backward compatibility creates baggage for legacy SaaS in AI era
“Having the older way of doing things actually serves as baggage because you have so many customers already locked into that approach. And because you have a lot of customers, anything you build has to be compatible with all of them. And so it just always inher…”
Zhang: Customer experience is one of few AI markets with PMF
“It's, in my opinion, one of the few markets right now that has true PMF with AI.”
Zhang: Enterprise human labor budgets are 10x larger than software spend
“Human labor is generally like an order of magnitude larger than software spend, like 10 X or more.”
Zhang: Decagon clients spend more on AI agents than CRM software
“And so if you look at a lot of our customers, for example, they're spending more on the AI agent Than the previous CRM software for support.”
Zhang: Fully deployed AI customer service achieves 60% to 80% resolution
“But I would say that, like, a good bar when you're fully up and running is, like, 6070, 80, in that range. And it just depends on, like, what sort of things the AI has access to. Like, if it's able to take a lot of actions and resolve a lot of things, then, of…”
Zhang: AI customer service resolution will reach 80% to 90% soon
“I think in three to five years, it'll probably creep towards like just consistently 80 to 90, right? Because you're able to just so easily allow the end users to capture more logic and teach you new things.”
Zhang: Current AI models possess sufficient reasoning for customer service
“The reasoning capabilities matter a little bit less. I would say the models as they stand today are generally good enough at solving most of the meaty inquiries in customer service. I think it's like, instead of reasoning, it's more about instruction following…”
Zhang: Very few enterprises build AI customer support fully in-house
“I would say from our experience so far, very few people will build it themselves.”
Zhang: Approximately 50% of Decagon's new code is AI generated
“Yeah, roughly there. The lines get blurred because oftentimes what will happen is like, hey, I'm sitting down to do this project. Let me just have the agent Take a stab at it first, and then you go in and like edit things and move things around. So there's a b…”
Stebbings: Hiring will be AI B2B's biggest challenge over 18 months
“The single biggest problem in AI B to B will be hiring for the next year, 18 months.”
Zhang: In-person work is significantly more productive than remote work
“It's just way more productive. My last company, we were remote, so because it was during COVID, and I think there's pros and cons of that, but I think overall, just personality-wise, it just comes down to the founder's personalities. Like, Ashwin and I are doi…”
Decagon grew from zero to eight-figure ARR in one year
“Last year we went from roughly zero to eight figures ARR and we raised towards the end of that.”
Decagon rejected funding offers up to 2x higher than $1.5B
“I will say we could have raised at a much higher valuation up to like 1.5 to two X.”
Zhang: Maximizing valuation risks turning healthy startups into zombie companies
“When you think about people that raised that huge valuations, I, I've known folks that have done this. You get to a point where you maybe are still doing well as a business, but the markets change and you're just not able to raise at that price with some healt…”
Zhang: Enterprise AI customer experience market will have multiple winners
“I think our space probably will have multiple winners, hopefully not too many, and hopefully we're one of them. I think it's just hard to say, like, why there would be a huge single winner here. The reason is that there's not that many network effects from bet…”
Stebbings: Manual AI model selection will disappear within a year
“It's like, for me, the other one is like choosing which model you run on. Are you kidding me? We're not going to do that in a year.”
Zhang: AI systems will learn from human examples rather than prompting
“I think there'll be more of a learning from examples that will happen. Cause if you just think about how humans learn, if you bring like a really good human support agent, they learn by shadowing and kind of seeing examples and so on. There's probably going to…”
Zhang: Decagon prioritizes raw clock speed over experience across all roles
“Technically we really value just like clock speed really, really highly. And clock speed, I just mean like how fast your brain works and how fast you can learn. And this is across all functions, right? Obviously it's important for engineering, but for Sales an…”
Stebbings: 20VC missed ElevenLabs' seed round by rushing for efficiency
“It was 250 K at twenty five million. We're a big fund. It's like one percent, but I should have done it and I would have done it if I'd spent more time on it and I hadn't tried to focus on efficiency and clock speed.”
Zhang: Product-first approaches outperform forward-deployed engineering in consistent enterprise software
“And so, I think our mistake was we over-indexed on that a little bit. Because I don't think that concept necessarily makes sense in all use cases. And in our use case, the types of things that people care about across different customers is generally pretty co…”
Zhang: Workplace wellness programs exacerbate stress by creating contrast with relaxation
“If you invest so much in the wellness stuff, it actually makes the stress worse because now there's this huge juxtaposition between like relaxing and stress. Well, instead, just like embrace it.”
Stebbings: Revolut's Nikolay Storonsky is the best founder he has interviewed
“Honestly, Jesse, I've interviewed a thousand of the biggest founders of our time. Nick is the best of all of them. Period. Hands down.”
Zhang: Valuation gap between OpenAI and Anthropic will shrink over time
“I don't think that long-term necessarily we'll always see a five X difference between open and anthropic. So I guess, yeah, maybe, maybe if the question is more like, what is spread trade? I think like that, that multiple will shrink. Over time.”
Zhang: Narrative that AI will transform every use case is overhyped
“The fact that there's this narrative that AI is just going to transform every single use case. I think what we've seen is that most use cases have not been transformed by AI, especially if you think about enterprise use cases and things like that, is that, yea…”
Zhang: Competitor Sierra relies on heavy, Salesforce-style configuration
“The approaches are just very different because of their leadership all coming from Salesforce. They've taken a very Salesforce-esque approach where it's, you know, the heavy configuration and sort of this longer lift to get going. And we're more of a, you know…”