Grit · 21 Sep 2026 · From the week of 21 September
Inside the AI Startup Making Customer Support 65% Cheaper
These are notes on the conversation, checked against its transcript. The episode itself has the full discussion.
In brief
Kleiner Perkins partner Joubin Mirzadegan interviews Jesse Zhang, co-founder and CEO of Decagon, an AI customer-support agent company that grew from an idea in 2023 to a $4.5 billion valuation and from roughly 50 to 500 employees in a year. Zhang covers hiring only when the pain was felt for the first 18 months, then hiring ahead at scale with bar-raiser interviews. He explains how Decagon got out of the idea maze in about three weeks using sales-style discovery, how preempted rounds and investor relationships were handled, and why he favours mid-career investors early. His central argument is that Decagon's differentiation against Sierra is a deployment model in which customers build, test and monitor the AI agent inside the product and own the results, as opposed to a Palantir-style forward-deployed engineering model. The two also trade notes on marketing hires, sales comp, co-founder fit, actionable culture values and keeping intensity sustainable.
For founders
- Zhang credits a roughly three-week idea maze to running founder discovery like a sales cycle, asking prospects about price, budget source, ROI justification and who is involved instead of asking whether they would use the product.
- Decagon hired only when it felt the pain for its first 18 months (to about 50–60 people), then switched to hiring ahead and 10x'd to 500 people, adding late-stage bar-raiser interviews for culture and intensity to protect the bar.
- Zhang argues no investor will help you find product-market fit, so early on he favours mid-career investors who have time and will prioritise you, and he tests investors' helpfulness before a raise by asking for introductions while they are trying to win the deal.
- Zhang says co-founder fit depends more on wanting the same things out of life, equal commitment and operating at the same level than on how long you have known each other.
- Zhang says culture values only work when they are actionable rules, such as giving feedback directly to the person, rather than abstract statements like valuing transparency.
For revenue leaders
- Decagon's founders did all sales until about $1M in revenue across roughly six customers, and only hired the first salespeople once there were already deals for them.
- Zhang says Decagon wins large enterprise deals against defaults like Salesforce and Google through a deployment model in which the customer builds, tests and monitors the AI agent in the product, positioned against Sierra's forward-deployed engineering approach.
- Decagon's first marketing hire came within the first 30 employees and was a product marketer; Zhang believes growth and performance marketers should come later, once marketing can be run more systematically.
- As Decagon scaled, its sales team shifted from creative self-starting generalists to more experienced reps who want to execute in a playbook organisation, while the company kept a high intelligence bar.
- Decagon's founders built early sales comp plans themselves by borrowing plans from sales leaders they knew, before hiring finance at around 30–40 people, and the plans changed a lot over time.
What was said 35, most useful first
Decagon added bar-raiser interviews at the end of its hiring process to keep the bar high while scaling, and most candidates pass them. Listen
Zhang says the hard part of hypergrowth hiring is not finding candidates, since many people want to work at AI companies, but keeping the bar high. Modelled on Revolut and Amazon, Decagon added a final-stage filter that selects mostly for culture and intensity, introduced roughly as it geared up to go from 50 to 500. Zhang and Ashwin personally met every hire until about 300 people. Long-tenured employees seen as good representatives of the culture now run these interviews. A majority pass, and Zhang says that if the last step cuts most people, something is probably wrong.
“You kind of have these filters that you add on to the interview process at the end, bar raisers or whatever you want to call them, where they're just mostly selecting for culture and intensity and so on.”
Decagon got through the idea maze in about three weeks by running founder discovery like an aggressive sales cycle. Listen
Avoiding a long idea maze was an explicit goal at founding. Zhang says most founders shy away from being aggressive because it feels too salesy, and default to soft questions like 'would you use this?' Instead, he and Ashwin asked how much the buyer would pay, where the budget would come from, how ROI is justified internally and who is involved. Zhang acknowledges luck played a part.
“they'll appreciate if you're asking more tough questions, like how much we pay for this, where's the budget coming from, like how do you justify ROI internally, like who are the people involved?”
No investor will help you find product-market fit, so early on he favours mid-career investors over senior GPs. Listen
From early investors Zhang expects three things: no negative impact, emotional support, and hustle on introductions to hires and customers. Since no investor can find PMF for you, he thinks mid-career investors who have the time and motivation to grind for you are the early-stage sweet spot. He sees very senior GPs as very good during scaling. He presents this as the approach Decagon took, not the only one; Mirzadegan said he took a signal-first approach instead.
“So I think optimizing for mid-career investors in the early stages, I think is like actually the sweet spot. Because they have the time and the sort of motivation to really grind for you.”
Decagon felt it had product-market fit at roughly $1M in revenue from about six customers, with the founders still doing all the selling. Listen
Zhang says there was no clear cliff, only a gradual scale-up. Around $1M from about six customers paying real contracts felt repeatable enough to signal demand. He calls customer support and coding the two big horizontal AI use cases, with law and some healthcare more vertical. He says they did not know this going in and that mapping out the market upfront is not wise because it is noisy.
“Six customers, a million in revenue. It's like, okay, well yeah, there's a lot of people that want this. People willing to pay us real contracts.”
After a certain point, raising mainly serves to reprice equity for recruiting, because candidates value equity by its dollar figure. Listen
Early rounds bring validation and more people around the table, but Zhang sees diminishing returns beyond that. If you are hiring very fast, a higher valuation helps because nearly every candidate looks at the dollar value of their equity, which he says they shouldn't.
“after a certain point, though, I think there's diminishing returns of that, and like raising mostly helps you just like reprice your equity.”
Zhang tests investors before a raise by asking for introductions and help while they are trying to win the deal, since they will never be more helpful than then. Listen
Zhang reasons that investors are maximally incentivised during the courtship phase, so if they are not helpful then, they won't be once on the cap table. Decagon did this for nearly every round and mostly responds to inbound investors rather than seeking them out. He believes Decagon may leverage its cap table more than any other company. He describes a self-reinforcing loop: when an investor's intro closes a deal, you tell them and their team, and they help more.
“no investor will ever be as helpful as during that phase where they're trying to close a deal. Cause like their sole focus, they're like super incentivized to help.”
Decagon's main differentiation is a deployment model in which customers build, test and monitor the AI agent in the product and own the results. Listen
Zhang describes Sierra's model as Palantir-like, relying on forward-deployed engineers who do the work for the customer. Decagon's founding thesis was that this model makes less sense as AI models improve, because agentic AI lets non-technical team members build more, as with Claude Code. Decagon still staffs white-glove people for enterprises, but its configuration is not code. Customers own the agent's results and the process. He credits this product approach for winning big deals despite default options from Salesforce and Google.
“can you build a product where the AI is fully in there, it gets built in the product, it's tested in the product, all the conversations that it's having, you can monitor in the product.”
Zhang disagrees with advice to only co-found with someone you've known for years; he puts shared life goals, equal commitment and similar operating level first. Listen
Zhang met Ashwin through mutual friends and began discussing a company at an Andreessen Horowitz offsite. He calls finding a co-founder luck-based, like finding a spouse: you need enough shots. When Mirzadegan described YC's advice to work with someone you've known for five years, Zhang said that if that is the prevailing advice he disagrees. Long friendships may make alignment more likely, but alignment is what matters.
“I think the thing that does matter a lot more are, hey, do you want the same things out of life? I think that's the most important thing. Are you equally committed to being a founder?”
Mirzadegan pushed back on his head of sales keeping the head of implementation on all pre-sales calls, because it stops her from hiring. Listen
At Roadrunner, the head of sales wanted the head of implementation on every call, including pre-sales, to put the company's best foot forward. Mirzadegan argued she needs to go hire, and that he himself had stepped back from selling despite possibly being the best seller, to avoid being the bottleneck on every customer. He accepted there may have been a short-term tax. Zhang added that recruiting feels like a short-term opportunity cost but skipping it creates much higher long-term costs.
“Was there a tax that we had to pay by like not putting our best foot forward in air quotes? Maybe. But like eventually, like what am I going to do? Just like be the bottleneck to every customer. It doesn't work.”
Listen to the episode Sales team, hiring & comp Link to this
Decagon hired only when it felt the pain for its first 18 months, including sales, and Zhang thinks in hindsight this may have been a mistake. Listen
For roughly the first 18 months and first 50–60 people, Decagon did not hire ahead. Zhang and co-founder Ashwin did all the sales, and the first salespeople were hired when there were already deals waiting for them. Zhang says that in hindsight, given how strong the market turned out to be, 'maybe' they should have hired ahead, but in the moment the stories of companies that over-hired and then had layoffs made them cautious.
“We only hired someone when we actually felt the pain of, okay, we need someone doing this. That went for sales as well. I mean, in the beginning, Ashwin and I were doing all the sales, and eventually when we hired our first salespeople, it was like we already had deals for them.”
At hypergrowth scale Decagon switched to hiring ahead because reactive hiring breaks down. Listen
Decagon roughly 10x'd its team in a year, from about 50 to about 500 people. Zhang says that at this pace you have to plan ahead, because you cannot reactively hire this many people without things breaking. The first roughly 100 people were not hired much ahead of need.
“Now, of course, we do hire ahead just because you have to. You have to plan out where you are. Because otherwise, you just cannot reactively hire this many people and things are going to break.”
If a final-stage interview rejects most candidates, something is probably wrong. Listen
Asked what share of candidates fail Decagon's bar-raiser step, Zhang did not know the exact number but said it is definitely a minority. His reasoning is that because it is the last step, a high cut rate there signals a problem.
“If you're cutting the majority of people, the last step to the process, that's probably something wrong.”
Decagon let its first strategic salespeople work remotely because they would be travelling to customers anyway. Listen
Decagon relaxed its in-office norm first for its strategic sellers. Zhang reasons that keeping all salespeople in one office defeats the purpose, since they are out meeting customers. They come into the office when they travel to it. Sales now has big hubs in New York and London, and the New York office, opened about nine months earlier, is already on its third space.
“Because we found people that we thought were really good, and we realized that these people are not- they're going to be traveling a lot anyways.”
Listen to the episode Sales team, hiring & comp Link to this
A single big open-plan office is very undervalued, and he picks rectangular floors so everyone can see each other. Listen
Decagon's current SF office is donut-shaped in a high-rise, and Zhang found that people on one side don't see people on the other side. The new office is one large rectangle, with a goal of 500 people in one space, and walls are being torn down to achieve it. The New York office is also a big rectangle.
“I actually feel very strongly having one big open space is like very, very undervalued.”
At hypergrowth Decagon feels like a new company every quarter or two, and first-time founders' naivety can help. Listen
Zhang says processes and team structure are reinvented roughly every one to two quarters and are built iteratively as the company grows. It is the first time he and Ashwin have run a company this size. He sees an upside to this: being naive lets them stay optimistic and try things from first principles.
“it does feel like kind of a new company every quarter or two, roughly.”
His previous startup had almost zero tactical relevance to running Decagon, and prior B2B SaaS experience might even teach bad habits. Listen
His prior company, a consumer app sold to Niantic, taught him personal lessons: he disliked being a solo founder, disliked working remotely, and prefers B2B. It also gave him intuition for spotting ideas that waste time. He says it taught him almost nothing about leading a team or running a company. He agreed with Mirzadegan that even B2B SaaS experience would not transfer much because 'everything's faster now.'
“But in terms of tactically, how to lead a team, how to run a company, almost zero learning.”
Zhang found solo founding hard because you are a single point of failure and lack someone with equal context to test ideas against. Listen
Zhang gives two reasons. The first is psychological: no matter the team size, only the founder has founder responsibility and context. The second is that ideas take much longer to resolve without someone of equal context and commitment to talk them through, because talking out loud eliminates bad ideas quickly. He notes you can now partly do this with Claude as a sounding board.
“Now you can actually do some of this with Claude, which is kind of interesting. You can use that as bounce ideas”
Decagon raised its seed round from Andreessen Horowitz before incorporating or having an idea, and most investors then advised against the idea it chose. Listen
Zhang says he and Ashwin decided to be co-founders and raised a small seed from a16z with zero ideas. When they settled on AI customer support, most of their investors and some very smart founders told them not to do it because it was too obvious. They ignored the advice. Zhang concludes that no one has the depth to judge your idea.
“most of our investors and some very smart founders we knew, we're like, no, don't do this. This is a dumb idea because it's too obvious.”
All of Decagon's rounds were preempted roughly six months apart, and Zhang still deliberately pushed raises off to avoid over-high valuations. Listen
Zhang says Decagon never ran a fundraise; each round came from an inbound term sheet, and the founders often delayed because they are conservative about raising at too high a valuation given market cycles. He says six months between rounds is too frequent all else equal, but in heavy growth you have to raise to keep momentum, and he warns against planning on that cadence.
“All the rounds were roughly, like, six months apart. Which again, it's not always going to be like that, so we shouldn't plan for that. But six months, I think, is a little bit too frequent.”
Zhang suggests raising small rounds more frequently can work as a recurring marketing event, citing Ramp. Listen
Agreeing that a raise is a marketing event, Zhang offers this as something you 'could do' rather than something Decagon does. He points to Ramp as doing it very well. Each raise gives you an announcement that many people are guaranteed to read, where you can place your messaging.
“I think like one thing you could do is just like raise small rounds more frequently. Like Ramp does this very well, for example.”
Decagon uses quarterly recaps to warn employees that the current hot market won't last. Listen
Zhang says the company tries to keep a steady mindset and not get too high in the highs. Many companies grew fast in the zero-interest era, struggled for a couple of years and are growing again, so growth is cyclical. After strong quarters, leadership tells everyone it won't always be this easy and they need to build toughness now.
“when we recap the quarter, we tell everyone, like, hey, obviously, it's great right now. Like, everyone should be ready, and that it's not always going to be so easy.”
Decagon's sales team shifted from creative generalists to more experienced reps who want a playbook to execute, while keeping a high intelligence bar. Listen
Zhang says generalists are most valuable when spinning up any function, and specialists come as functions get defined. Decagon's original sales team was very smart, creative and self-starting, and Zhang says you cannot find hundreds of people like that. Newer hires include more experienced people who want to join a playbook organisation and execute, which he calls normal.
“Like some of them are much more experienced. They want to kind of come into like a playbook organization and just execute. And I think that's very normal.”
Listen to the episode Sales team, hiring & comp Link to this
Decagon wins large deals against incumbent platforms only because its product is superior. Listen
In a market with defaults like Salesforce and Google, and a competitor led by a well-connected former Salesforce CEO, Zhang says Decagon cannot rely on relationships. He attributes its big-deal wins to a product that empowers customers to build and own their AI.
“and yet we're winning a lot of these big deals. And so the only way we can do that is through product.”
Zhang names marketing as the function he most misunderstood and says Decagon probably hired for it too slowly. Listen
Zhang says marketing is several distinct disciplines: product marketing, growth and demand gen, and brand and social. None is fully systematic, and there is some art to each. He says Decagon could probably have hired ahead in marketing, but it was hard to identify the pain point that would trigger the hire. It is the function he had to learn the most about.
“marketing, I would say, one, I think we hired it a little bit slow. I think that goes back to one of those, like the original question of like, could we have hired ahead there? Like probably yes.”
Decagon's first marketing hire was a product marketer within the first 30 employees, and Zhang thinks product marketing suits an amorphous early company. Listen
The hire, who is now VP of marketing and came from Glean, has done very well, though Zhang notes they don't know the counterfactual. His view is that product marketing is amorphous and so fits a young company that is itself amorphous, where a smart generalist can flex to needs. He would hire growth and performance marketers only once marketing can be run more systematically at scale.
“I think the thing with product marketing is because it's more amorphous, that actually like fits better in your early company because your early company is like very amorphous.”
The hardest part of founder time allocation is building the first bench of people who do your work better than you. Listen
He says deciding where to spend time is a constant effort of self-reflection but not especially hard. What is hard is evaluating early hires when you may have no clue what you are doing yourself, because you then have to judge whether someone will be world-class at it. He says this gets easier once a team starts forming.
“I think the hard part is like forming the initial bench of like just like teammates that can do a lot of things that you would be doing and do it better than you.”
Zhang seeks advice as data points from founders about a year ahead, rather than opinions, and says these peers get scarcer as you scale. Listen
Zhang says founders roughly a year ahead in the same environment have the most to offer because everything is fresh. He cites Harvey as being exactly a year older and gracious with its time. The useful advice is 'here's how we did it', for example how a company ran a sales leader search. He thinks giving opinions on others' situations is dangerous because you don't know them in depth.
“at our current stage, there's fewer and fewer companies that have been at this stage in this environment a couple months ago or six months ago.”
Decagon's founders set early sales comp plans themselves by borrowing from sales leaders they knew, before a first finance hire at 30–40 people. Listen
Zhang says they did not find comp plans hard. They asked sales leaders they knew and used those plans. The plans have changed a lot over time, which he describes as normal. The first finance hire came at roughly 30–40 employees.
“Yeah, sales comp plans. We just asked like some sales leaders that we knew. And we just used them. They have changed a lot over time”
Listen to the episode Sales team, hiring & comp Link to this
Zhang and Ashwin push back hard on territorial politics by redirecting people to their own work. Listen
Zhang says politics starts as soon as there is a team and is impossible to avoid entirely. He and Ashwin stamp it out aggressively. When someone argues they should own a thing someone else owns, the response is to ask whether they shouldn't be focused on their own responsibilities.
“Like if someone is coming to you and saying like, hey, like I should own this other thing. And like someone else is owning it right now. Shouldn't you be worried about your own stuff right now?”
Culture values must be actionable rules like 'give feedback directly' rather than abstract statements like 'we value transparency'. Listen
Zhang describes this as a lesson Decagon was coached on. He says he has no opinion on whether direct feedback is the right tenet, but it is actionable. A company should build a list of concrete behaviours like that, because 'we value transparency' is much weaker.
“It's like that is much weaker than like, okay, anytime you want to give feedback to someone, you tell them directly. Yeah. I think you have to build a list of actionable things like that.”
Decagon tried catered meals so a 200-plus person company would eat at the same time again; people liked the dynamic but disliked the food. Listen
Zhang says the value of catering is that everyone eats together, recreating the early days. Ordering individually means meals arrive at different times. The experiment was well received for the dynamic, and the company plans to find better food. He also tells employees to remember each stage as 'the good old days' because it won't last.
“one of the values of doing catering is that everyone eats at the same time. And so, yeah, so we ran an experiment with catering and people liked it”
Zhang keeps one day off a week to sustain intensity, and treats Sunday as an optional deep-work day with no expectation to respond. Listen
Zhang and Ashwin are in the office every day except Saturday. Zhang says working seven days leaves him less energised the following week, so a day off keeps up intensity on the other days, though he says there is no one answer for everyone. Sundays are a meeting-free pseudo workday. He says there is zero expectation that anyone comes in or responds, and the number of people who come in on Sundays has stayed roughly what it was at 20–30 employees.
“if I'm working seven days a week, I feel a lot less energized the following week. So having a day, at least for my natural cadence, is pretty important to keep up the intensity in the other days.”
Roadrunner's offer acceptance rate fell as it scaled, because it now competes for different candidates with less ownership to offer. Listen
Mirzadegan runs Roadrunner, which Kleiner Perkins incubated, at about 35–40 people. He says the company used to brag to its board about a very high hiring acceptance rate, but it is now much lower. Scaling puts it in more competitive fights for great candidates, with a different profile and different economics. The first 10 hires had ultimate ownership, while later hires join as part of a team.
“We used to brag about how our, tell the board, like, you know, our hiring acceptance rate is so high. Yeah. And now it's like much lower. And it's because we're trying to scale the company.”
Starting a company will stay easy even in a downturn, which keeps recruiting hard for entrepreneurial talent. Listen
Decagon has had very few departures, but a few people left to start companies. Many of its employees have been founders or want to be. Zhang says that with so many companies and a low barrier to founding, people leave to start companies even without an idea. He doubts a downturn would change this much. He hopes, as Mirzadegan suggested, that a downturn might ease recruiting.
“I think even when there's a downturn, it'll still be pretty easy to start a company.”
The zero-to-one phase is by far the hardest stage, and the feeling stays roughly constant once you are scaling. Listen
Zhang compares growth to getting taller: no single day feels different. The one big shift he has felt is moving from the idea maze to post-PMF scaling. His previous company spent about three years in that first phase, when he was a new grad. In hindsight, he says, more upfront thinking would have shown some ideas were bad.
“So every stage has its challenges, but the zero to one phase is by far the hardest.”