The Best Little Unicorn in Texas: Jasper Was Winning the AI Race—Then ChatGPT Blew Up the Whole Game

Still, Rogenmoser felt a sense of urgency. In a matter of months, he had become an emerging star in the exploding field of generative AI. Jasper had raised $125 million in October, reaching a $1.5 billion valuation only 18 months after launch. It had ramped up to 100,000 customers, three-quarters of whom paid $80 or more each month for access to Jasper’s premium suite of AI-powered writing templates. Its revenue, which The Information reported as $30 million in 2021, was on track to double by the end of 2022, according to the company. OpenAI, by contrast, has earnings in the low tens of millions of dollars, according to reporting by The Information—revenue that largely comes from startups like Jasper licensing its APIs. (OpenAI expects these partnerships to drive $200 million in business in 2023, according to reporting from Reuters. A spokesperson for OpenAI declined to comment directly about revenues.)

Before ChatGPT came along, Jasper was drawing unconditional raves from customers: One user told me it performed basic writing tasks as well as or better than an entry-level employee, easily justifying the $100 she paid each month for the company’s “boss mode” product, which offered access to long-form writing templates and up to 50,000 generated words. Another customer, a web developer, told me he had stopped hiring freelancers to write web copy and started using Jasper instead. “The potential is limitless,” he said.

Through last month, Jasper’s growth matched that potential. In January 2022, the company had just nine employees. By November, it had over 160, including more than 70 new engineers and designers tasked with building more, and better, customized templates. Most of Jasper’s 60 current templates are tailored to the needs of marketers, who use the service to generate blog posts, press releases, social media captions and ad copy. There are also templates to brainstorm product names, generate viral tweets or distill long articles into brief bullet points. Those looking for something less defined have an option to go free form, prompting Jasper to write anything from scratch.

The company is also working on features to make Jasper accessible outside its own platform, like a Chrome extension that brings AI-customized copy into other parts of the web. “We basically want to supercharge every text box on the internet,” Rogenmoser told me in November. “Not even just on the internet. Every word that you write, I would love to be there assisting.”

Then ChatGPT arrived. It became obvious within days that this was not merely one step forward in the advancement of generative AI, but potentially a black swan event. “Nobody was expecting ChatGPT to be this good, this fast and this free,” said Chris Frantz, co-founder of Snazzy AI, a content marketing startup also built on GPT-3.

When startups like Jasper and Snazzy launched in early 2021, GPT-3 still required high-precision prompting to produce a salient result. That made the copywriting startups more essential, because they bridged the gap between the immensity of an AI language model and the utilitarian needs of an end user. ChatGPT turned that relationship on its head. With ChatGPT, “you can say, ‘Generate a blog post about X,’ and it’s great. That’s all you have to do,” said Frantz. “So the defensibility around the prompting”—the entire value proposition of prompt-based startups like Jasper and Snazzy—“that’s suddenly gone. That’s vamoose.”

Plus, who would pay $80 a month for Jasper when they could get ChatGPT for free?

The moment “really focused everybody,” Rogenmoser told me. Within days, he had mobilized a team to come up with a response plan, which involved Jasper building its own chat feature—one that, like ChatGPT, can respond to edits conversationally. (The feature launched on December 20.) The team started thinking about building more enterprise features for its existing clients at large companies like Airbnb, HarperCollins and IBM. Jasper also finalized the acquisition of Outwrite, a browser extension that improves grammar and fluency in applications like Google Docs and Microsoft Word. That came with another perk: Outright already had a million users, 10 times Jasper’s user base.

In a matter of days, Rogenmoser had gone from presumptive winner in the text-based AI startup space to embattled CEO defending his turf. Jasper had been early to catch the wave of generative AI and had so far proven capable of riding it. Now the company had to prove it could hold on for the long haul.

---

Two days before the surprise release of ChatGPT on November 30, I paid a visit to Rogenmoser at Jasper’s office on the western outskirts of Austin. From the outside, the building looked like a generic office park, but inside, Jasper HQ reminded me of the startup offices that started popping up in San Francisco’s SoMA district a decade ago. It had foosball and Ping-Pong tables, snacks in glass jars and cafeteria-style tables where employees eat lunch together every day. One conference room had been turned into a golf simulator. Rogenmoser had declined to give himself his own space, instead working Zuckerberg style, perched at a desk in the middle of the open-floor office.

I found the CEO sitting there in front of an ultrawide curved monitor. Rogenmoser is extremely tall—6 feet, 8 inches—but his towering height is balanced by the aw-shucks demeanor of a kid from Topeka, Kan., who’s thinking of running for homecoming king. He is also uncommonly candid for the CEO of a billion-dollar company. When I asked about his choice to study marketing as a student at Kansas State University, Rogenmoser confessed that he was more focused on being “the best frat guy” on campus than on his academics. He said he skipped many of his classes and occasionally cheated on his exams.

Rogenmoser and I spoke in a small conference room with whiteboard walls, where someone had artfully depicted the Teenage Mutant Ninja Turtles shouting, “Jasper is radical!” Unlike Altman, who had started OpenAI out of a deep conviction that AI would unlock an age of “abundance” and benefit all of humanity, Rogenmoser just wanted to get rich—or, achieve a “certain level of financial freedom,” as he put it. He looked back on the eight years of stumbles that preceded Jasper’s breakout as an exercise in perseverance.

His initial entrepreneurial ideas, like a software product for fraternity recruitment, played to his life experiences. (“It was a bad idea,” he admitted.) Later, he and two friends from college, J.P. Morgan (yes, his real name) and Chris Hull, started a small marketing agency, which offered copywriting services for website content, Facebook ads and search engine optimization. Neither Rogenmoser nor his friends had those skills, so they outsourced the work to contractors and took a cut of the profit.

Morgan and Hull eventually became Rogenmoser’s co-founders at Jasper. Morgan, who taught himself how to code, is the only one of the three with technical skills; he built Jasper’s back end and serves as chief technology officer. Hull oversees people and operations as chief operating officer. The three have known they wanted to work together for as long as they’ve known each other. They all liked to golf and shared many of the same midwestern values—between them, they have 10 kids.

After launching the marketing agency, they build an online course to teach other people about marketing, which garnered them the email addresses of thousands of people working in advertising, communications and sales. To help sell their course, Morgan built a little pop-up window that appeared on their website to increase conversions. It worked surprisingly well—enough so they turned that into its own business, called Proof.

Proof was accepted into Y Combinator in 2018, later raising $2 million and growing to 15 employees. But the startup struggled to scale. “We could never figure out how to take it from a feature into a platform,” Rogenmoser told me. At the end of 2020, he let half of Proof’s employees go, promising to rehire them when the company figured out how to fix the business.

That was around the time Rogenmoser started seeing people talking about GPT-3 on Twitter. OpenAI had introduced the model in the summer of 2020, and AI insiders were practically salivating about what it represented. GPT-3 was built on a transformer model, a type of neural network that infers relationships between enormous amounts of data scraped from the internet. Transformer models offer “a much broader set of capabilities around communicating in natural language,” said Michael Wooldridge, who teaches computer science at Oxford University and is the author of “A Brief History of Artificial Intelligence.” OpenAI had already built two models, but GPT-3 was notable for the enormity of its training set. “It’s 500 billion words that it’s been trained on—basically, the entirety of the World Wide Web,” said Wooldridge.

In late 2020, OpenAI set up a small, private beta for developers to see how they might use the technology in the real world. The beta was extremely limited, but Rogenmoser used his Y Combinator connections to get access that December. (Altman was previously a partner at Y Combinator, though he and Rogenmoser had never met.) Only a handful of developer teams had access to GPT-3, giving Rogenmoser and his buddies a significant head start in developing a product.

Once they started playing around with GPT-3, they knew exactly what they wanted to do with it: reach the same customers they had with Proof, but offer them far more functionality. Morgan spun up a prototype, which the team named Conversation AI, and Rogenmoser took it to a few corporate customers to see what they thought. “I’d been selling this personalization software for two years, and it was just crickets,” he told me. “With this, people were falling out of their chairs.”

Conversation AI launched in January 2021. (The team later changed the product’s name to Jarvis, after Tony Stark’s AI butler in “Iron Man,” then changed it again to Jasper after they received a cease-and-desist notice from Marvel.) Access to the platform cost $29 a month, with eight templates at launch. It wasn’t the first to market—other startups, like Copy AI, had already started harnessing GPT-3 for copywriters—but thanks to his work on Proof, Rogenmoser had a vast mailing list of marketing professionals to tap as potential customers.

Selling it was as easy as showing people how to use it. Thomas Laffont, co-founder and senior managing director at Coatue Management, described his initial demo of Jasper as a “magical experience”—one that, like seeing the first iPhone, convinced him the world was about to change dramatically. Coatue had already made investments in other AI companies, but Laffont loved Jasper’s team, which he said combined “complete scrappiness and big vision.” (It helped that Jasper was also generating significant revenue.)

In October, Coatue joined the $85 million Series A, along with Insight Partners, Bessemer Venture Partners and others. It made Jasper among the fastest startups in history to reach unicorn status—and an outlier in a year of depressed funding.

“They’ve been handed a huge opportunity,” Laffont told me. “They have the lead. There’s a clear product-market fit. They’ve curated a great set of investors with a lot of different expertise.” They also had customers, thousands of them, eager to exploit the magic of generative AI. All signs were pointing up and to the right. And then, out of nowhere, OpenAI entered the chat.

---

For half a century, futurists have imagined a world where advances in technology lead to a complete merging of man and machine. In this world, intelligent machines might effectively evolve the Homo sapiens species, just as the use of fire led to a leap in human brainpower. Computer scientist J.C.R. Licklider described this as “man-computer symbiosis” in a now-famous paper from 1960. Rather than computers supplanting humanity, Licklider imagined their relationship as a partnership, where “men will set the goals, formulate the hypotheses, determine the criteria, and perform the evaluations. Computing machines will do the routinizable work.”

We have never seemed closer to this reality. The development of GPT-3 has already enabled a dizzying array of new possibilities: It can instantly produce hundreds of lines of code, perfect Shakespearean sonnets or pose suggestions to solving geopolitical conflict. It can write news articles, draft academic dissertations and distill the themes of novels better than SparkNotes. The implications of a text generator as powerful as GPT-3 are far-reaching, and the responsibility to safeguard its users largely falls on its maker, OpenAI. But the constellation of startups that will build the infrastructure for this new world will also face decisions about who AI is for and how it should be used.

Early on, Jasper marketed itself as a tool to supercharge copywriters’ productivity. “All of our marketing is about how we’re going to make you successful,” Rogenmoser told me. “On the hero’s journey, our customers are the hero.” But it didn’t take long for users to come up with slightly less heroic use cases for the product. Some tapped Jasper to help with things like completing homework assignments or writing school papers. “It’s a great way to ensure that your paper will be original with no plagiarism, and it can write on any topic in any format you specify in advance,” one blogger wrote.

When I asked Rogenmoser how Jasper would deal with his product’s potential to upend education, Rogenmoser seemed surprised by the question. “It would kill me if we made everyone dumb,” he said. “That would be really bad.”

It wasn’t so much that products like Jasper herald the end of thinking, he suggested. There are other ways to assess learning, like oral presentations or Socratic debates. Maybe school essays wouldn’t survive the AI revolution, but Rogenmoser wasn’t so sure that was a bad thing. “I would say people 20 years ago probably thought learning to spell perfectly was [valuable] learning—that if you can’t spell something right, you can’t live a good life,” he said. The advent of spellcheck programs challenged that assumption, but didn’t replace thinking or learning outright. “I am definitely not for cheating,” he added. “But things are changing. I think schools are going to have to figure this out.”

There were other off-label uses I wanted to know about. One of Jasper’s customer success agents told me she occasionally sees people trying to use it to generate erotica and other explicitly sexual content. Rogenmoser said those use cases violated Jasper’s content policy, which also prohibited defamatory, discriminatory and unlawful content.

While we were sitting in the conference room with the Ninja Turtles, I told Rogenmoser I felt wary of how people might use his product. The history of emerging technology suggested it was only a matter of time before Jasper introduced fake news, racist conspiracies or misogynistic bias into a professional marketing campaign. Rogenmoser, on the other hand, struck me as hyperoptimistic—a characteristic that, I realized, he shared with OpenAI’s Altman. Rogenmoser said he was “not wired” for cynicism, and rattled off a few of the great things that generative AI promised, like helping people with dyslexia communicate better.

I must have looked dubious, because he paused. “I do worry about being a company that’s not actually doing any good,” Rogenmoser said. “I would probably put Facebook in that camp. And it’s like, after all that work, after [Mark Zuckerberg] devoting his whole life to it, it’s almost better if it didn't exist... I don’t want to accidentally be there in a couple of years, thinking, ‘Crap, why did we do this at all?’”

---

If you put Rogenmoser and Altman side by side, you’d notice more than a few similarities. Both are white men in their mid-30s who grew up in the American Midwest. Both are professed idealists, startup CEOs and Tesla owners. And both have become rich (or in Altman’s case, richer) in the AI gold rush. But where Altman has become a whale in the open ocean of AI, Rogenmoser is more like a remora, the ray-finned fish that attach themselves to cetaceans and feed off their debris. OpenAI needs partners like Jasper to pay the bills, but not nearly as much as startups like Jasper need OpenAI.

Joanne Chen, a partner at Foundation Capital, told me Jasper’s reliance on OpenAI was one of the “biggest risks” she flagged before investing in its seed round in 2021 and its Series A in October. “But since then,” she said, “I think that risk has gone way down.” Chen was encouraged by the number of other large-language models being developed by companies like Meta Platforms, Amazon, Nvidia and Google. And Jasper has already begun to incorporate other open-source models, like GPT-J, in its product. “These models are becoming better and better, and we’re getting to a point where I think we’re going to have lots of different options.”

For Jasper’s customers, their decision to stick with the service will largely depend on whether it works better than the competition. Ayrton Mendoza, a designer and web developer for Postweld, an online marketing agency, told me Jasper still outperformed ChatGPT for copywriting tasks because of the way Jasper’s templates were designed. Mendoza used to hire freelancers on websites like Fiverr to write marketing copy for the websites he makes, but the process was costly and time-consuming. “It saves so much time,” he said of Jasper. “You can write a whole article in minutes.”

Mendoza’s favorite feature is Jasper’s ability to alter tone, making an email sound more cheerful or injecting humor into ad copy. “I can make those adjustments there and then, instead of having to message back and forth [with a freelancer], like, ‘Hey, can you word it this way?’ or ‘Can you use a different tone?’” he said. “I’ve also tried saying, ‘Write this like Joe Rogan,’ or ‘Write this like Elon Musk.’” The results have been amusing—and sometimes shockingly accurate.

Other customers said Jasper’s product worked well, but not well enough to justify its price. “I used Jasper for a couple months, but I was able to get much better output from ChatGPT within 10 minutes,” said Grant Tucker, founder of creative studio Chromaspring. Tucker found Jasper’s pre-set templates confining and preferred the conversational interface of ChatGPT, which made revisions easier. Recently he asked ChatGPT to summarize a campaign he had done with Adidas. “It gave me this rough, short blurb about it, and I was able to respond with instructions: Make this longer. Don’t talk about how close the relationship was,” he said. Overall, he was pleased with the experience. With Jasper, he said, “you’re paying for a lot of bloat versus the actual tool. ChatGPT has no bells and whistles, but for the vast majority of us, that’s all we need.”

Rogenmoser thinks that’s fine. He sees Jasper’s future more as an enterprise platform, like Atlassian or Slack, that companies might pay to integrate into their workflow. “ChatGPT can do a lot, but it lacks, and will probably continue to lack, all of the functionality that bigger businesses want,” he said. “We see ourselves as the company that’s going to help bring AI to businesses.”

Other startups are hoping to do that, too—bringing to market a constellation of new enterprise products that pipe GPT-3 into emails, text messages and web browsing. Optimistic to the core, Rogenmoser didn’t feel too concerned about that competition either. “We’ve gotten here by being quick, by being scrappy,” he told me. “That’s how we’re going to earn our spot six months from now.”