Frank Slootman is the only CEO in history to take three enterprise software companies public. Data Domain, then ServiceNow, then Snowflake. At their peaks those three were worth more than $200 billion combined. He took ServiceNow from roughly $100M in revenue through an IPO to $1.4B. He joined Snowflake out of retirement in April 2019 when it was worth $4 billion and led it to the largest software IPO ever recorded. He has two economics degrees from Erasmus University Rotterdam.
That was the archetype for twenty years, or at least one archetype. Founder builds the thing, board brings in the operator, operator installs the sales machine, sales machine builds the billion-dollar company.
One detail cuts against the way this story usually gets told. Slootman never carried a bag. He joined Compuware in 1993 as a product manager, ran UNIFACE in Amsterdam as GM, then ran the EcoSystems division in Campbell. From 2000 to 2003 he was SVP of Products at Borland, running the engineering and product management function. That was the job he held the year before his first CEO seat. There is no VP of Sales or CRO anywhere on his resume. The sales reputation came from how he ran companies, not from how he got there.
Marc Benioff fits the archetype better: 13 years at Oracle across sales, marketing and product, Rookie of the Year at 23, the youngest VP in company history. Benioff also started as a programmer. He founded Liberty Software at 15 writing and selling Atari games, and he wrote assembly code in Apple’s Macintosh division during college.
Now look at who runs the fastest-growing B2B and AI companies today. Databricks. Replit. Harvey. Fireworks AI. Sierra. Decagon. Abridge. OpenEvidence. Count how many came up through sales.
Zero.
The five numbers this post rests on:
- 82 out of the 100 fastest-growing AI-native startups are led by a technical CEO. In the 2013 Unicorn Club, the comparable number was 49%. Across all founders, 86% of the AI cohort is technical versus 59% a decade ago.
- Databricks and Snowflake ran the controlled experiment for us. Same category, same era. Snowflake hired a professional outside CEO twice. Databricks promoted a Berkeley professor with no business experience. Databricks is now at $190B growing over 80%.
- Snowflake then replaced Slootman with an engineer and product revenue growth went from 26% to 34%. Sridhar Ramaswamy’s own explanation involves rebuilding the sales organization, which is the part everyone skips.
- The third category is the domain expert. Harvey is run by a securities litigator at $11B. Abridge is run by a practicing cardiologist at $5.3B.
- Bill McDermott is delivering 24.5% growth at $15.8B of subscription revenue and the stock spent most of 2026 down roughly a third. The sales-guy CEO still produces the number. The market has stopped paying for it.
#1. Eight companies, eight CEOs, no sellers
Start with the roster.
- Databricks, $190B. Ali Ghodsi has a PhD in distributed computing from KTH in Sweden, was an assistant professor there, went to Berkeley’s AMPLab as a visiting scholar, co-created Apache Spark, and is still an adjunct professor at Berkeley. Six of the seven Databricks co-founders have computer science PhDs.
- Fireworks AI, $17.5B. Lin Qiao spent seven years at Meta as a Senior Director of Engineering leading 300-plus engineers building PyTorch, the framework most of the industry now trains on. Five of the seven co-founders were PyTorch core contributors. The company crossed $1B in annualized revenue this year, up fivefold, and moved from 15 trillion to more than 40 trillion tokens a day. Jensen Huang called her company the TSMC of AI factories.
- Sierra, $15.8B. Bret Taylor co-created Google Maps, was CTO of Facebook, founded Quip, and chairs OpenAI’s board. He was also co-CEO of Salesforce, which is the closest thing on this list to a commercial resume, and he got that job as an engineer who built things first.
- Replit, $9B. Amjad Masad wrote security software for internet cafes in Amman as a teenager, was a founding engineer at Codecademy, and ran JavaScript infrastructure at Facebook. Replit raised $400M in March 2026 and is targeting $1B in ARR.
- Decagon, $4.5B. Jesse Zhang is an Olympiad mathematician who studied CS at Harvard and worked at Google, Hudson River Trading and Citadel before selling his first company to Niantic.
- Harvey, $11B. Winston Weinberg was a securities and antitrust litigator at O’Melveny & Myers. His co-founder Gabe Pereyra was a research scientist at Google DeepMind and Meta with an Oxford PhD track funded by DeepMind. Harvey went from $100M ARR in August to $190M in January and is used by more than half of the 100 largest US law firms.
- OpenEvidence, $12B. Daniel Nadler has a Harvard PhD and previously founded Kensho, which S&P Global bought in 2018. His co-founder Zack Ziegler is a machine learning researcher.
- Abridge, $5.3B. Shiv Rao is a practicing cardiologist at UPMC who still sees patients one week a month. He co-founded the company with Zach Lipton, a Carnegie Mellon professor whose machine learning program Rao had previously funded as an investor.
There are non-technical CEOs in this cohort and some are doing well. Brendan Foody at Mercor is a Thiel Fellow who went from $100M to $500M annualized in six months and is worth $10B. Victor Riparbelli at Synthesia has a computer science and business degree and recruited two professors as co-founders. May Habib at Writer graduated with high honors in economics from Harvard.
None of the three is a career seller either. Nobody in the top tier of this cohort got the job on a resume that reads: carried a bag, ran a region, ran worldwide sales, got handed a company.
#2. 82 out of 100, versus 49% a decade ago
Leonis Capital indexed more than 10,000 AI startups from 2022 to 2025 and pulled out the 100 fastest-growing AI-native companies. Across that set, 82 of 100 are led by a technical CEO and 208 of 241 founders (86%) are technical. The comparison set is Aileen Lee’s original Unicorn Club, where 49% of companies had a technical CEO and 59% of founders were technical.
Research pedigree moved the most. 40% of AI 100 founders have a research background versus 12% in the Unicorn Club, and 58% of the companies have at least one co-founder who came out of research. Berkeley PhDs, DeepMind and OpenAI alumni, Olympiad medalists. That founder type barely existed in the last cycle.
They are also younger. Median age at founding is 29 in the AI cohort versus 34 in the last one. The single most common age is 26 or 27. These founders went from a lab straight to a company instead of spending a decade climbing.
Leonis is a venture firm that invests in technical founders, so discount the framing accordingly. The underlying dataset is open-sourced and the gaps are 30 points wide.
#3. Databricks kept the professor. Snowflake hired the operator. Twice.
Two companies started in the same category at the same time and made opposite calls on this exact question.
In January 2016 Databricks was in trouble. It had raised around $174M at roughly a $1B valuation and had almost no revenue. AWS and Cloudera were absorbing Spark into their own products. Founding CEO Ion Stoica agreed to step aside and return to his Berkeley professorship.
The obvious move was to bring in a seasoned Silicon Valley executive, which is exactly what Snowflake did on its way to a $33B IPO. Instead the co-founders pushed for Ali Ghodsi, then VP of Engineering, a career academic who had never run a business. Ben Horowitz, the company’s first VC and a board member, thought that made no sense. His words, roughly: swap out one founder-professor for another? The board compromised on a one-year trial.
Horowitz has since called Ghodsi the best CEO in a16z’s portfolio, which spans hundreds of companies.
Databricks announced a $5B round at a $190B valuation in August 2026, growing more than 80% year over year, with over 1,000 customers spending more than $1M a year and more than 100 spending over $10M. Ghodsi has declined to take it public.
Snowflake is a very good company. The company that kept the professor is four times larger and growing twice as fast.
#4. Snowflake swapped the operator for an engineer and growth went from 26% to 34%
On February 28, 2024, Snowflake announced Slootman was retiring and Sridhar Ramaswamy was taking over. The stock dropped 20% and wiped out $15 billion in a day. The market’s read was obvious: the guy who knew how to sell was leaving.
Ramaswamy is a computer scientist who ran Google’s advertising business from roughly $1.6B to over $100B and then founded Neeva, an AI search startup Snowflake acquired.
The trajectory since:
Q1 FY27 was the strongest sequential dollar growth in company history. Net revenue retention went 124% to 125% to 126%. Remaining performance obligations hit $9.77B, up 42%. The full-year FY27 product revenue guide was raised to $5.84B. The stock jumped 36% on the print and gained more than 50% over five days.
Asked what drove it, Ramaswamy credited two years of grinding away at making the sales organization an AI-native organization, alongside the product work.
So the engineer CEO spent his first two years rebuilding sales. He treated it as an engineering problem, which is a different thing from treating it as somebody else’s problem.
One flag: Snowflake reports Q2 FY27 on August 27, so the 34% figure above is one day from being superseded. Guidance was 30%.
#5. A litigator at $11B and a cardiologist at $5.3B
The “engineers won” framing misses a third group entirely.
Winston Weinberg was a junior lawyer at O’Melveny & Myers doing exactly the document-heavy grunt work that Harvey now automates. He is not an engineer. His co-founder is. Harvey went from $100M to $190M ARR in five months, grew from about 300 employees to more than 1,000 in a year, and raised at $11B.
Shiv Rao is a cardiologist who majored in history at Carnegie Mellon and studied film theory. He founded Abridge in 2018 after leading UPMC’s provider-facing investment portfolio, and he still practices medicine one week a month. Abridge is live in more than 250 health systems.
Daniel Nadler has a Harvard PhD and had already built and sold Kensho before starting OpenEvidence, now at $12B.
What these three share is first-hand knowledge of a workflow that a model had just become good enough to do. Weinberg knew which parts of litigation prep were mechanical. Rao knew which parts of a patient encounter turn into documentation burden. A great sales leader cannot acquire that in a quarter, and neither can a great researcher.
Leonis found that 9 of 13 vertical AI founders in their set had direct sector experience.
#6. Bill McDermott is putting up 24.5% at $15.8B and the stock is down a third
The counterexample gets a full section, because the sales-guy CEO is not failing on the numbers.
Bill McDermott owned a deli at 16, spent 17 years at Xerox rising to President of the U.S. Major Account Organization and SVP/GM of Xerox Business Systems, was president of Gartner, ran worldwide sales operations at Siebel, then spent 17 years at SAP where he became sole CEO. Every one of those roles was a sales role or a sales-org P&L. He is the genuine article.
ServiceNow’s Q2 2026, reported July 22: subscription revenue of $3,877M, up 24.5% year over year and 23% in constant currency, beating the high end of guidance by 150 basis points. Total revenue of $3,987M. cRPO of $13.20B up 21%, RPO of $29.0B up 21%. Operating margin of 29.5%, three points above guidance. AI annual contract value crossed $1B in the quarter, on track for $1.5B by year end. 123 transactions over $1M in net new ACV, up nearly 40%. Full-year subscription guidance raised to $15.77B. Rule of 56.
That is an extraordinary quarter at that scale. McDermott’s line to Fortune: “We are who we said we were.”
And the stock entered that day down roughly a third for the year and about 50% off its 52-week high, having fallen another 6.5% that same session on a report that OpenAI was building an enterprise product. It jumped as much as 7% after hours on the print.
The playbook still works. It produces 24% growth and Rule of 56. What changed is the price the market pays for that outcome. Slootman’s version of the job was to take a good product and multiply it through distribution, and investors have moved their underwriting from the multiplier to the thing being multiplied.
McDermott is also the only person in this post with a purely commercial path to the corner office. Slootman ran engineering and product management at Borland. Benioff shipped Atari games at 15 and wrote Macintosh assembly at Apple before he ever sold anything at Oracle. Ghodsi co-created Spark. Ramaswamy built Google Ads. The category we call the sales-guy CEO has roughly one clean member in it.
#7. Technical CEOs pivot in 12 months. Non-technical CEOs take 27.
The mechanism here is timing, not intelligence.
Two-thirds of the AI 100 pivoted at least once, versus 54% of the Unicorn Club. Researcher-led teams pivoted in a median of 12 months versus 18 for non-researcher teams. And companies with technical CEOs pivoted in 12 months while companies with non-technical CEOs took more than twice as long, at 27 months.
Fifteen months is a long time in a market where model capability jumps every few months and entire product categories appear and disappear on the back of a release.
Cursor’s founders were building AI CAD software for mechanical engineers. They got early access to GPT-4, tested it, and within minutes recognized it was extraordinary at coding. They abandoned months of work and rebuilt as a coding assistant. Making that call in minutes requires someone who can evaluate a model directly instead of commissioning a study.
Slootman’s own stated philosophy is situational leadership over playbooks, so he would likely agree with the underlying point. In 2014 the situation you had to read was the buyer. In 2026 you also have to read the model, and reading the model is a technical act.
The sequencing changed too. More than 80% of the AI 100 launched with self-serve onboarding. Sales still happens, after adoption, formalizing demand that already exists. A CEO whose core skill is manufacturing demand from scratch is optimizing the second step.
#8. Every one of these companies is hiring the sales org Slootman would have built
Here is the case against everything above.
The sales machine is still getting built at all of these companies. It gets built later, and the person running it reports to the CEO instead of being the CEO.
Amjad Masad wrote publicly that he thought he hated sales culture, and that by the end of this year more than half of Replit will be salespeople. Anthropic went from fewer than 10 startup sellers to 150-plus in 18 months. Lovable hired a CRO out of Klaviyo at around $200M ARR. OpenAI’s go-to-market went from 50 people to 700-plus in 18 months. I wrote a whole post on this a week ago and the version from 2021 still holds: eventually, almost everyone has a sales team.
Three more reasons to hold this loosely.
The cohort is two to four years old. Slootman never joined a company pre-product-market-fit. He took over Data Domain, ServiceNow and Snowflake after the product worked, which means the comparison is stage-mismatched. The real test of the AI cohort’s leadership is what happens at $500M to $1B in ARR with a competitor, a churn problem and a bad year, and almost none of them have been there.
Half the numbers in this post are private marks set in the hottest funding environment in history. Leonis notes that many of these companies still run poor or negative gross margins because inference costs scale with usage. A $17.5B round prices a forecast.
Selection bias runs through the whole dataset. We are counting companies that broke out. Nobody counted the technical founders who could not sell anything and quietly died.
Three decisions this should change
If you are a founder deciding whether to hand over the company: the 12-versus-27-month gap says a CEO who cannot evaluate a model release directly runs 15 months late to the pivot that matters. Bringing in an operator does not close that gap, it widens it. Get close enough to the product to read a model release yourself, or give a co-founder who can the authority to set direction.
If you are a board running a CEO transition: Databricks in 2016 and Snowflake in 2024 made the same call and both worked. Promote someone who understands the product deeply, then make them own the revenue number. Ghodsi turned around a company with no revenue. Ramaswamy spent two years rebuilding a sales org. Being technical did not excuse either of them from the commercial problem.
If you are a great sales leader: the job is still enormous. McDermott runs a $15.8B subscription business at 24.5% growth and Rule of 56, and Snowflake’s Chris Degnan took the company from zero to $3.5B as CRO across four different CEOs. The seat you are being hired into is CRO or President, and the CEO you report to will be someone who can read a model card. If you want that CEO seat, take the path Slootman and Benioff actually took rather than the one we remember them taking. Slootman ran engineering and PM at Borland the year before Data Domain. Benioff spent his Oracle years in product development as well as sales. Go own a product line for three years.
One last data point. Frank Slootman is an angel investor in Fireworks AI, the company founded by the woman who ran PyTorch at Meta. He is not fighting this. He is buying it.
