Christina Cacioppo on Vanta's AI Strategy

Here is a puzzle that has nothing to do with bad management. A software company grows faster than it ever has, keeps its margins fat, and still watches investors cut its valuation to a quarter of what it used to be. That is the position Christina Cacioppo found herself in over the past year, running Vanta, the compliance and security platform she started eight years ago (02:06). Vanta now serves roughly sixteen thousand customers across five products (02:56), connects to four hundred other tools to pull in security data (03:55), and by any historical measure is a very good business. The market, for the moment, disagrees. Cacioppo calls the feeling of living through that gap cognitive dissonance, and her explanation of it is the clearest thing in this conversation about what it is actually like to run a mature software company inside an AI gold rush.

Her argument has a number attached to it. She believes an 80 percent growth business with 80 percent margins will, over a long enough horizon, return more cash to its owners than a 160 percent growth business with 30 percent margins. Today, she says, the market pays roughly three to four times the valuation multiple for the second kind of business (06:04).

"I think in the fullness of time, a eighty percent growth business with eighty percent margins will actually be better than over the same time period, run it out long enough, a hundred and sixty percent growth business with thirty percent margins." — Christina Cacioppo [06:04]

She is not claiming the market is wrong forever, only that it is wrong right now, and that she has to operate inside that wrongness anyway, because the people she hires are reading the same headlines she is. If a rival offers an engineer a story about a trillion-dollar outcome, Vanta has to compete for that engineer's attention even while Cacioppo thinks the story is often fantasy.

The clearest place this tension shows up is pricing. Vanta's product increasingly runs on AI agents that act on a customer's behalf: offboarding an employee's accounts, flagging a lapsed control, asking permission to fix something and then fixing it. Every one of those actions burns tokens, the units of text a language model processes, and tokens cost real money. The obvious move is to charge for them. Cacioppo has decided not to, at least not yet. Her reasoning is specific: if Vanta marked up a token from Anthropic's Claude model by four times and passed that cost to customers, those customers would quickly notice they were paying a premium for something they could evaluate directly against the model itself (10:02).

"But anyone who is marking up Claude tokens 4X and trying to pass that off to their users, I think is just saying that." — Christina Cacioppo [10:26]

So Vanta eats the cost, accepting a lower margin in order to get as many customers as possible into agent-driven workflows, which currently cover about a quarter of its customer base, with a stated goal of three quarters (10:46). The logic is sequencing: win the usage and the retention first, and treat pricing as a problem to solve later, once the product's value is proven rather than argued for in an abstract markup debate. It is a bet that current margin compression is a phase, not a new identity.

The Wizard Nobody Could Admit They Wanted to Replace

The most concrete story in the conversation is a small one about onboarding software, and it says more about organizational psychology than any abstract claim about AI could. Vanta wanted to rebuild how large customers get their existing security data into the product. The traditional approach is what Cacioppo calls an onboarding wizard: four steps, a progress bar, maybe some confetti at the end. She told her team repeatedly to abandon that pattern and instead build something closer to a blank canvas, where a customer could drag in messy files and screenshots and let a language model make sense of them (19:40). Her team, by her account talented and motivated, kept bringing her wizards anyway (20:01).

When she finally asked why, the answer was not a technical limitation. It was fear. The team did not believe they were allowed to fail, and the wizard was the design they already knew would work, while the AI-driven canvas was a gamble (20:22). Cacioppo's response was to grant explicit permission to fail and tell them to commit fully.

"Burn the boats and just make this thing work." — Christina Cacioppo [20:22]

The team shipped the canvas. It works well, by Cacioppo's account. The lesson she draws is not really about large language models at all. It is about how often an organization's hesitation around a new technology is actually a hesitation about being blamed for an experiment that does not pan out, dressed up as a technical judgment.

A related discipline she insists on for herself and other leaders: build something with the current generation of AI models personally, at least every two weeks, because models change fast enough that secondhand impressions go stale (18:16). The point is to have a visceral, not theoretical, sense of what a model can and cannot do, so that when someone claims a single agent can replace an entire team, a founder can ask where that claim is true and where it falls apart.

Paying People More Than They're Worth, On Purpose

Talent competition is where the growth-versus-margin tension becomes personal. Engineers at AI labs and fast-growing startups are being recruited with stories of valuations climbing five times in a year. Cacioppo's counter is not to match the story but to be honest about Vanta's numbers, good and bad, and let candidates judge for themselves, a habit she frames as a company value: do what it says on the tin (22:09).

On pay itself, she has become more aggressive, not less. The highest earners at Vanta now make roughly ten times what their equivalent would have earned five years ago (24:12). Her stated philosophy is to keep trying to pay the very best people enough and to accept she will never fully succeed.

"My comp philosophy is you never pay the top people enough. You should always try, and you will never hit it." — Christina Cacioppo [24:12]

She is explicit that this is not about a skill category like sales or engineering. It is about what she calls high agency: a person with enough context about the company and its customers to make reasonable decisions quickly, including some bad ones, and then correct course. That, she argues, is worth far more than its salary line suggests, and the gap between what such people are paid and what they generate is widening, not narrowing.

Cacioppo is just as direct about what she thinks is overvalued in the current moment. She is skeptical of companies post-training their own AI models, meaning adjusting a model's internal weights using a company's own data, a process that is expensive and usually unnecessary. Inference providers themselves, she says, will tell a customer the real reason to post-train is to cut cost, not to improve quality, for most use cases (29:09). The appeal she sees instead is marketing: being able to tell investors and customers that a company trained its own model, whether or not it does anything measurably better.

A stray detail from her travels complicates the usual assumption that AI geopolitics is mostly about national pride. She learned that many software companies in Paris have switched to using Chinese AI models, not out of any ideological preference but because of a business continuity scare: when a change in U.S. policy briefly cut off access to an American model for non-U.S. users, French founders concluded they could not safely build a business on infrastructure that might vanish by decree (33:13). It is a small story with a large implication: AI infrastructure choices are starting to look like supply chain decisions, not technology preferences.

Security itself, in Cacioppo's telling, is heading toward a world where defenders stop ranking which vulnerabilities to fix first and simply fix everything as fast as it is found, because AI-driven attackers do not get tired and will chain together many small, individually low-risk flaws into something serious (35:44). "If you're fixing things at machine speed," she says, "you don't have to prioritize" (36:27). It is a vision of security as a constant, automated back-and-forth rather than a quarterly audit.

The episode closes, fittingly, on a story about losing a billboard on Highway 101, the one Vanta had become known for after a marketing stunt went viral on Reddit (50:06). There was never a real multi-year contract, just an informal placement order, and a competitor eventually paid more and took the spot out from under them (52:52). Cacioppo offered to double the asking price and still lost, because the new tenant was flush enough not to need the money (53:28). She joked that Vanta had probably just lost some market value the moment the sign came down. It is a small, almost silly story, but it rhymes with the larger one: in this market, being first, being clever, even being loved by the internet for a week, buys less certainty than people assume. The thing that holds, by her own account, is simply shipping the next useful thing before someone else does.