How I Invest with David Weisburd Sep 18, 2026 1h 15m 53m saved
With Marty Kausas, co-founder and CEO of Pylon, the customer-support software company backed by Andreessen Horowitz and General Catalyst, which he started with two engineer friends six days before ChatGPT launched
Pylon picked its market before it picked its product. Marty Kausas and his two cofounders looked at every public software company worth more than $10 billion, asked what they had in common, and concluded the answer was the size of the market rather than the quality of the team.
Most founder interviews open with a problem the founder personally hit. Kausas said that framing is usually dishonest, and that his own reason for running the company is that he enjoys it.
"Our reason is we just love the game."
Kausas started Pylon six days before ChatGPT was released, raised from Andreessen Horowitz and General Catalyst, and has since moved the company's internal target from $1 billion of revenue to a trillion-dollar market value. He now describes his competition as Anthropic rather than Zendesk.
The full interview is covered here so you can skip it. 75 minutes of audio, 22 minutes of reading.
Here are the 19 lessons that matter.
Key Takeaways
Support staff are already spending about $1,000 a person a month on Claude, which is the budget Pylon is going after
Pylon studied public software companies and found market size, not team quality, was the differentiator
Salesforce's biggest revenue line is Service Cloud at $10B a year, not the CRM it is known for
The internal goal moved from $1B of revenue to a $1T market value, because that is what "generational" now means
He says he overestimated the incumbents — Zendesk and Salesforce are too slow to execute, not too strong
Full AI replacement is not what is selling; the 50% of tickets an agent closes were the easy half
"Almost every repeated Claude prompt or skill is a company waiting to be built" is his test for new products
He interviews candidates by doing the market-size math in front of them when a rival offers double the equity
The hiring question he says almost nobody asks is whether you would want this person as your manager
Category creation, not product marketing, is what he says the best software companies actually do
His biggest mistake was people management, which he calls the next thing after product-market fit
1. Fun and Adventure
Asked why he started Pylon, Kausas said he did not start it to change the world, and that most founders are not straight about their own reasons.
He thinks the standard founder story is usually untrue
My guess is a lot of founders are not honest with why they start companies, and we are.
Marty Kausas
The alternatives do not hold up on inspection, he said. If the motive is money, there are easier ways to make it. If the motive is a problem you hit at a previous job, ten years of grinding on it reads as disingenuous — and he was blunt that he is not personally passionate about customer support.
The actual motive
Our reason is we just love the game.
Marty Kausas
Fun and adventure is the reason, he said, and it works as fuel because it does not run out.
2. Market Beats the Team
The motive is the "why." The "what" is a target: Kausas described the company as a board game whose win condition was a billion dollars of revenue inside ten years. To work out how, they looked at who had already done it.
They studied public software companies worth more than $10 billion at the time and found there were not many. What separated them was one variable.
Market size dominated every other input
And what we realized was if you take the best founders and the best team and put them in a small market, the market wins.
Marty Kausas
The reverse also holds, he said: weak founders in a great market still do well. A great team in a great market is where a generational company comes from.
The constraint is arithmetic
You just can't grow out of that small market.
Marty Kausas
The host put the math plainly: 20% of a $1 billion market is a $200 million business, and 20% of a $10 billion market is a $2 billion one. Kausas added a caveat about how markets are measured — the early crypto market looked small, but the real market was money movement, which is not small. What he wanted was a market that was already large, with something new happening inside it.
3. Why Customer Support
The search for a large market was deliberately lazy, by his own description: find the biggest company in business software and go after its biggest revenue line.
Salesforce's largest business is not the one it is known for
And their biggest revenue driver, interestingly, is not their CRM, which is what they're known for, but it's actually their customer support platform, which is called Service Cloud.
Marty Kausas
Service Cloud produces about $10 billion of revenue a year, he said.
That decided the category
That's how we picked customer support as our starting category.
Marty Kausas
Selling the same product as an incumbent does not work, so they needed a change in the market to attach to. They found it by sending LinkedIn messages to people working in customer support asking what was new, what they cared about and which metrics mattered. One pattern came back: business-to-business companies were starting to run support through shared Slack and Microsoft Teams channels with their customers, and no existing support system could work outside email. Pylon's first product bridged Zendesk or Service Cloud into those chat platforms. Customer complaints about that bridge are what led to building a full support platform for business-to-business companies.
4. Recruiting on Market Size
The host's framing was that every chief executive says hiring is the most important thing and almost none of them think hard about the pitch to employees, while the best recruiters, in his list Elon Musk, Alex Karp and Palmer Luckey, are all polarizing in a specific way.
Kausas said the fun-and-adventure line lands well with investors and candidates precisely because nobody expects it, but that it cannot stand alone.
The thing a founder owes the company first
I think the greatest gift a founder and CEO can give to their company is product market fit.
Marty Kausas
Without it there is no potential energy for a new hire to convert. With it, and with a revenue target the company has already worked backwards from, the pitch has no ambiguity in it: here is the goal, the market, the incumbent and the arithmetic.
When a candidate has a competing offer, he does not argue about the equity.
He runs the comparison in the room
So for example, let's say we have an offer out and someone else is offering double the equity, I will go, I will do the math in front of them.
Marty Kausas
The comparison he draws is between category sizes. Customer support is a $40 billion software category, and the agent-based version will be larger. A search-optimization company, by contrast, competes with incumbents worth about $1 billion in total market value — not revenue. Against Service Cloud's $10 billion of revenue, he said, that is two to three orders of magnitude of difference. The two companies he concedes are hard to argue with on market are Anthropic and OpenAI.
5. Ambition Costs the Same
Before starting, the founders considered bootstrapping or picking a less competitive market. They dropped both on one observation: whatever they chose, they were going to put all their time into it.
The effort is fixed, so the target should be high
And so you may as well go for the most ambitious company you can.
Marty Kausas
Ambition also makes other things easier, he said, hiring first among them. His argument to candidates is that nobody gets many companies to go all in on, and going all in on a company aiming at $10 million of revenue and a sale is not interesting.
6. From $1B to $1T
Pylon started six days before ChatGPT was released. A billion dollars of revenue looked ambitious then. This year the founders revisited it.
The definition of generational moved
And generational now means actually you get to a trillion dollar market cap.
Marty Kausas
So they went back to the market, the positioning and the pricing, and concluded that a fully agent-based version of the product could multiply the revenue per customer by ten.
The unit price is the mechanism
So if we're charging a $100 per unit right now, actually in the future, we're gonna be charging a thousand.
Marty Kausas
His comparison is the software era: a company like Salesforce charged about $100 for a seat. What is happening now, he said, is that a support team member spends roughly a thousand dollars a month on Claude, for work that the seat used to cover.
7. Competing With Anthropic
The conclusion Kausas draws from that spending is a change of competitor. Salesforce cannot build the features its customers are assembling for themselves, on his read, because it is too slow.
The competitive set moved up a layer
So our competition moves from, hey, we used to compete with these support platforms to now we're actually competing with Anthropic for a very specific job function.
Marty Kausas
Everyone is using general-purpose models for work they are not optimized for, he said, which is expensive and slow. Pylon's aim is to move that spend from the model provider to itself, at an order of magnitude more than the old software line item.
The host, an Anthropic investor since its Series C, said he never believed a single general model would replace verticalized business software, and that the way companies buy software, handle compliance and execute makes it impossible.
8. Prompts as Companies
Kausas's version of the same point is a rule for finding products.
His test for a new company
Almost every repeated Claude prompt or skill is a company waiting to be built.
Marty Kausas
The host compared it to the old Craigslist diagram where each category became a company — Airbnb, Thumbtack, Tinder. Kausas gave a case from inside Pylon. The company's outbound callers were told to paste each call transcript into Claude for coaching, which worked, and which he then realized was three steps in two disconnected tools. Coaching belongs inside the calling software.
What the model is actually doing
And so I think Claude is filling the gaps because people haven't caught up on the product side and built the right thing yet.
Marty Kausas
Salesforce should be able to capture that revenue, he said, and cannot.
The incumbent's advantages are real and not enough
They have all the advantage right now in terms of distribution, in terms of brand, in terms of relationships, but they just can't execute on the building.
Marty Kausas
The host called it the innovator's dilemma and located it in hiring: the people a large company employs are not the people it employed in its first years, with less upside and less tolerance for risk, which he said makes it behavioral rather than technological.
9. He Overestimated Zendesk
Asked about the assumption he got wrong at the start, Kausas named one.
The mistake was fear of the incumbent
I overestimated the incumbents.
Marty Kausas
He used to assume a newly hired support executive would default to buying Zendesk, because that is what people do.
The reason the default does not hold
They're just so slow. That's the problem. They're just too slow.
Marty Kausas
A motivated team that is native to this generation of tools can move around them, he said, and a founder who executes should expect to be faster rather than slower. The host's counterfactual was that replacing the whole Zendesk team with a startup team on new equity terms would probably work, and that no company can do it.
10. AI Augments, Not Replaces
On the claim that AI is replacing people, Kausas was direct.
The replacement narrative does not match the products that sell
One of the narratives in the market has been AI is replacing humans completely, and I think that's false.
Marty Kausas
His evidence is the companies that have worked. Anthropic and OpenAI are augmentation tools. Cursor, which he called the only application company in AI that has really taken off, has not replaced software engineers and would have failed if it had tried full replacement. His test for the state of the technology is a small one: "until Claude can write a good LinkedIn post, I don't wanna hear about AGI."
The category he uses as the cautionary example is AI sales development.
The product nobody can show him in use
And somehow, I still can't find anyone who's using AI SDRs.
Marty Kausas
He said he would use them if they worked and has been asking for a case study without finding one. In customer support the same arithmetic applies, and it is unflattering to the deflection numbers vendors quote.
Automating half the tickets removes far less than half the work
Those 50% were the ones that were the easiest to answer, because we already had a knowledge based article about it.
Marty Kausas
The escalated remainder is where most of the work sits. Pylon sells deflection for the simple tickets because customers want it, and concentrates on the escalations.
11. Going Agentic, Defined
Kausas has a name for the change in how work gets done. The old sequence in support was to receive a ticket, route it, investigate, escalate and write the reply. The new one is that an agent does that and the person reviews it.
The definition
And so that transition from doing the work to now orchestrating agents that do the work, we call going agentic.
Marty Kausas
By the time a case reaches a person it has gathered the context and asked the questions, and the decision is to approve the work or send it back for more. Coding went the same way, he said: review, write and test became instruct and delegate.
12. The Last Mile Problem
Asked whether self-driving is the same shape of problem, Kausas said it is. Autonomous trucking could handle most of the drive and still needed someone to unload the truck.
Partial autonomy still requires the person present
The person still needs to sit in the car the whole time.
Marty Kausas
Keeping a driver with a steering wheel is what lets a company ship most of the benefit without waiting for the last few percent, he said, and it is easier to improve from there than to hold a product back until it is complete. Pylon's own AI sits in the same place: "90% of the time the answer's right, but there's cost when it's wrong," so a person checks both the answer and the reasoning.
Asked why, he offered two explanations rather than one. Part of it may be missing context — most of what a company knows is spoken and never written down anywhere a model could read it. Part of it is the model: he described long-running jobs where the system loses track of what he asked for two hours earlier. He cited Sam Altman on the related point that a human arriving for a job comes with two decades of other people's investment in them already paid for, where an agent starts closer to zero.
13. Intensity vs Intelligence
That line of thought produced an idea about how models get chosen.
Model selection may end up looking like recruiting
I wonder if a lot of model selection in the future will be more like an interview where you're looking at a resume of the work you've done
Marty Kausas
Today the competition is a handful of benchmarks, he said, which measure whether a model thinks well rather than whether it sells well. A model better at selling a product would need evidence and case studies rather than a score. He drew the parallel to his own sales hiring, where a candidate can be clearly intelligent and still lack the trait the job needs.
The dimension nobody benchmarks
Anthropic is probably optimizing for intelligence.
Marty Kausas
Intensity is what somebody chasing a number needs, he said, and it is not what the benchmarks measure.
14. Where Ambition Comes From
The host offered three predictors of ambition from hundreds of conversations with founders, the first being heavy trauma in childhood, then immigration or repeated moves, and then playing sports, individual or team. Kausas said the academic version did not apply to him.
He was not a driven student
I was a solid A minus student always.
Marty Kausas
He said he could not tell the difference between an A and an A-minus and did not care. His observation about the students who were pushed hard is that the pressure stops at college admission, and many of them arrive burnt out with the goal already met. His own version came later. An undergraduate girlfriend at Purdue told him he was not ambitious enough and not good enough, which he calls his tiger-girlfriend moment. He had a software-engineering job lined up at Airbnb and was about to out-earn his parents.
The ceiling moved at the moment he was becoming an adult
But suddenly the ceiling's raised, right as I'm graduating.
Marty Kausas
He said he was already ambitious and that this was fuel for it. The host's addition was that children raised only to get top grades end up fragile, because a career requires being bad at things first — he offered his own attempt at learning golf as the example.
15. Seriousness Doesn't Help
Kausas said he initially undervalued one of his two cofounders because the man did not present as serious, and calls that a mistake.
The trait does not predict what he expected
Turns out that seriousness does not correlate.
Marty Kausas
Sometimes the correlation runs the other way, he said. Being able to laugh about a setback is anti-fragile, and being too serious is what would have stopped him admitting the real motive for starting the company. He now asks new founders why they are doing it, specifically to hear whether the answer is honest, and said most of them fail that.
His example came from Y Combinator. Paul Graham told them that in office hours with founders who were pivoting, "the founders would have to go through all their serious ideas before they got to the good ideas" — the credible revenue plans came out first, and honesty arrived only once they were exhausted.
The case he tells is Twitch. Asked at the end of one of those sessions whether there was anything else, one of the founders said he had always wanted to put a camera on his head and livestream his life. Everyone found it funny and told him to do it. He became a local figure in San Francisco, and the company noticed people mainly wanted to watch him play games.
Removing the seriousness is what widens the search
And because you're able to remove the seriousness, you can try things that no one else would.
Marty Kausas
16. The Cofounder Contract
The host said the most underestimated predictor of whether a team succeeds is whether the founders like each other, because founders who merely respect each other pivot once and then conclude they gave it a fair try.
Kausas and his two cofounders were friends first — he had lived with one for three years, and the other two met at Caltech. What he says has to be aligned is three things: the motive, the target, and the working style.
Their name for the style
And we call ourselves internally happy grinders.
Marty Kausas
He described it as optimism and togetherness combined with high competence, and low ego. None of the three cared who was chief executive, and for the first two years the company did not tell anyone who held the title. Outsiders trying to work it out got answers about which part of the company each of them focused on instead.
How decisions actually get made
We will just debate heavily until we find the answer.
Marty Kausas
If a session does not produce an answer, that means there is not enough information yet, so they come back to it. He compared it to steel sharpening steel and said the volume alarmed the company's first employees, who thought the founders were falling out. The host's test for a good version of that process is that the answer, once found, looks obvious to everyone in the room.
17. Would You Work for Them
The hiring criterion Kausas says is most underrated is whether people like the candidate, and he turns it into one question.
The question he asks in every loop
The question is just, do you want to work with this person?
Marty Kausas
Most interviewers do not ask it, and new joiners often do not know they are allowed to raise it. He asked it of himself repeatedly while hiring a general manager for Pylon's new European operation: would he want to work for this person, and could they hire him. He kept going until the answer was yes.
Why proximity makes it non-negotiable
We're a five days a week in person culture.
Marty Kausas
Asked whether someone initially unlikable can win people over by being good at the job, he said no — they "end up being a wrecking ball to, like, relationships and culture." Pylon has interviewed candidates who were clearly very capable and lacked the other side, and passed. His reasoning connects back to the motive: an unpleasant colleague makes people not want to show up, which takes the rest with it.
18. Create the Category
The host raised Goodhart's Law — that what gets measured is what gets optimized, and that this is a fallacy, because an unmeasurable effect on culture is still an effect. Kausas said he had fallen into exactly that trap with marketing, where he required every team member's work to map to a pipeline number.
What has actually worked for Pylon is LinkedIn content, which is the hardest thing to attribute.
The evidence is indirect and still conclusive
All we know is that a lot of people book demos with Pylon, and they're like, I see you guys all over LinkedIn.
Marty Kausas
This is the first quarter he has accepted that some of the spending cannot be measured. The host's contribution was David Ogilvy's line that "90% of all advertising is wasted," and that you cannot tell which 90%.
The thing Kausas wants to spend that unmeasurable budget on is a category.
What he says the best software companies actually sell
The best companies usually don't promote their product, they promote an idea, or a way of working, or a new strategy.
Marty Kausas
His examples run through four mechanisms. HubSpot coined "inbound," a word that did not exist before search made it possible, then became the product attached to it.
The pattern in one line
They created the category.
Marty Kausas
Jira attached itself to Agile and Scrum, which were already spreading, and became the tool that implements them. Vanta attached itself to a certification that already existed and was growing: "And the more they push SOC two, the more Vanta benefits." Gainsight's founder pushed customer success as a category and the company became its product. Salesforce's real campaign, he argued, was not about customer relationship management, which already existed, but about software delivered as a service.
Pylon has done it twice. Slack-based support was already happening and lacked a product.
The term they coined and then heard back
And so we invented this term, B2B support, and then suddenly we started hearing people say it back to us, which was exciting.
Marty Kausas
The third attempt is "agentic support," which he defines as the same move from doing work to directing agents that do it. Customers are already assembling a version of it out of Claude, prompts and skills; the argument is that nobody should have to rebuild the same thing repeatedly.
19. Managing People Badly
Asked for his biggest mistake in three years, Kausas named people management, and put it directly after product-market fit in importance.
He had no experience of it
I went into this role never having managed anyone.
Marty Kausas
He called himself horrendous at the start, still working as an individual contributor and dropping into problems reactively. The specific failure was assuming other people wanted what he wanted. He and his cofounders are the same type — technical, indifferent to titles and to praise.
The correction
Actually, people want to be treated very differently than you might want to treat yourself.
Marty Kausas
Some people want to be told they did a good job, which he said was foreign to him. Exceptional people need different management, and being able to manage more kinds of them is what stops the founder being the bottleneck. He now uses a skill-against-will grid: a junior hire should be high will and low skill, and the fix is training; a capable person losing motivation is a will problem with a different lever. He also named the limit at this stage — there is no time to train someone reporting to a founder, so a skill gap usually ends in a departure.
Firing is less costly than managers fear
The moment you fire someone, usually everything gets better right away.
Marty Kausas
First-time managers worry about who absorbs the work and what the team will think; in his experience the top performers absorb it and the team's reaction is that it should have happened sooner. His broader claim is that good product-market fit with bad management decays anyway, because strong people have other options.
Bonus Insights
Niche podcasts are the underrated channel
The host's recommendation was the very small, very specific show — a podcast with a thousand chief technology officers or chief information officers in its audience, reporting 175 listeners an episode. Kausas said customer support has almost no thought leaders, unlike sales or marketing, which is why LinkedIn is Pylon's main channel and why podcasts have not been repeatable for the company. He said Pylon would sponsor people in the category if they existed.
And there is no conference either
The support conferences that exist run to about a hundred people, he said, and he sees the same faces at each one, while the population of support leaders is clearly much larger. He named customer workshops and AI support workshops as the format he thinks the category wants, because people know they want to use AI and want to hear from each other about how.
Clay's community as the model
Clay did a really good job of having a Slack community early on where so many people were excited because it was so new.
Marty Kausas
The demand built the community, he said, and the company's contribution was giving it an official place to happen. He contrasted it with the Zendesk Slack community, which he described as dead because everybody already knows how to use Zendesk.
The host's use of an existing schema
The host described taking the same idea to an extreme with an intern who had done very well academically at New York University: he split the summer into semesters, gave her progress reports and talked about her grade, on the reasoning that she was demonstrably good at that particular game.
Product-market fit covers a lot of internal damage
Kausas named Google and Airbnb as companies he has heard described as internally bad early on and successful anyway. On Google he relayed accounts of an early period with no managers and no direction, where teams built the same things and nothing was prioritized, while the search business was strong enough that none of it mattered and outsiders were trying to copy the management practices. On Airbnb he pointed at the founder's own public account of a company decaying internally until a reset during the pandemic.
Compounding is where the host puts his own time
The host's argument for focus was that a podcast is a network-effect business: each guest brings views and the views bring guests. He is adding a conference for chief information officers on the same logic. The things that compound are the hardest to start, he said, which is why they are worth the concentration of resources.
Kausas's bottom line is that the market decides the outcome, so the job is to pick a large one, name the new way of working inside it before anyone else does, and then not lose the company to bad management on the way.
Products, Companies & Tools Mentioned
Pylon (The customer-support platform he runs; internal target moved from $1B of revenue to a $1T market value, with unit pricing he expects to go from about $100 to about $1,000)
Salesforce (Service Cloud is its largest revenue line at about $10B a year, which is what made customer support the target category; he says its agent product is not real yet)
Anthropic (Named as Pylon's actual competitor for support budget, and as a company optimizing for intelligence rather than intensity)
Zendesk (The incumbent he says he overestimated, and whose user community he calls dead because nothing about it is new)
Cursor (His example of the only AI application company that has really taken off, and of augmentation rather than replacement)
LinkedIn (Both the research channel that found Pylon's first wedge and its single most effective marketing effort, which he says is also the least measurable)
Slack and Microsoft Teams (The shared channels that moved business-to-business support off email, which was Pylon's original opening)
HubSpot, Jira, Vanta and Gainsight (His four examples of category creation: coining inbound, attaching to Agile, operationalizing a security certification, and pushing customer success)
Clay (The community he says was built well, because the demand existed and the company gave it a place)
Y Combinator (Where Pylon went through the program, and the source of the Paul Graham observation about serious ideas coming out first)
Twitch (The company that came out of the unserious idea at the end of an office hour)
Airbnb and Google (His two examples of companies carried by product-market fit through bad internal management)
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: