With clients like Uber, Salesforce, Inuit and Qualcomm, the customer roster of A.I. startup Writer reads like a who’s who of the Fortune 500. The San Francisco-based company offers a suite of A.I. tools designed to deploy agents deploy agents and integrate the technology into business workflows.
Writer was co-founded in 2020 by May Habib, who previously launched a predecessor to the startup in 2015 called Qordoba,, a company focused on natural learning processing (NLP). Despite her relatively recent pivot to generative A.I., her experience with the underlying technology spans well over a decade. “I’ve been doing this for 15 years,” she told Observer in an interview in May.
Writer’s “end-to-end agent builder platform” is powered by its proprietary Palmyra family of models. Unlike rivals with sky-high training costs, Writer touts its cost-efficiency—some of its newer models have been trained for as little as $700,000.
That approach has not only attracted marquee clients but also significant investor interest. In November 2024, Writer raised $200 million in a Series C round that valued the company at $1.9 billion. As CEO, Habib reportedly holds a 15 percent stake—worth an estimated $285 million.
And she believes the company is just getting started. “It feels like agentic A.I. is going to eat so much of what has been the labor of people,” she said, adding that it’s difficult not to get excited about “watching what happens as folks automate much of what they were doing before.”
Observer caught up with Habib to discuss the startup’s role in the ever-competitive field of A.I. agents. The following conversation has been edited for length and clarity.
Observer: When did you first become aware of and interested in A.I.?
May Habib: I feel like the folks who are in A.I. right now are two different species—pre-ChatGPT species and the post-ChatGPT species.
We worked in NLP back when we were building aligned datasets for statistical machine translation models. So even predating real deep learning techniques. And even when we raised our Series A for Writer, we talked about machine learning and we talked about NLP and talked about transformers, but we never used the words “A.I.” because it was kind of taboo.
You have a wide range of clients. How much do they pay for these tools, and what are some of your favorite use case examples?
We sell to the Fortune 500 companies—these guys are spending $100 million a year. About $2 million or $3 million is going to Writer and that’s the highest ROI part of their investment. We understand that they need to go try everything, and what we really try to do furiously is continue to prove that this stuff works at scale.
Depending on the vertical and the industry, the use cases are pretty widely different. In the payer space services, like your health care insurers, some of my favorite use cases that we’ve got in production have to do with helping members really fully leverage their plans.
In pharmaceuticals, some of the best use cases are helping the sales people in the field really better understand who they’re selling to and be prepared for conversations, which can take hours of preparation because you’re reading such dense research material.
In the [consumer packaged goods] space, it’s been really exciting watching folks in real time optimize the listings of their products on Amazon.com and Walmart.com to really be able to sell more product.
In retail, folks are using Writer and agents to just combine sentiment and customer feedback in a really, really specific way and then propose action against it.
So just killer use cases that help drive really high revenue growth that is not possible without tools like Writer. We helped launch Airbnb experiences—that new product that they launched was launched with Writer. We wrote 37,000 pages, and they just simply would not have done that at that scale without that technology.
What would you say to those that fear A.I. will take away jobs or replace human labor in certain aspects?
People resisted writing as a technology thousands of years ago because it would make our memories weaker. It’s just par for the human course. We are going to be hand wringing about anything that changes the way people make a living or the way that we live, and there’s no question this is going to fundamentally change the way that we live, but it’s just never been more exciting.
None of our 300 customers are cutting jobs here. Everybody’s got mountains of A.I. related roles that they can’t find people to do. I think we’re going to be at just this incredible productivity, at full employment, but we do have to make sure that we are building really accessible experiences and join the market in really equitable ways. But there’s no question this is an overwhelmingly positive thing for humanity.
A recent report from Writer that analyzed A.I. adoption in the workplace found that two-thirds of leaders said it caused internal tension. What do you think is the right way to navigate adoption?
I see the tension that this is causing every single day. So many organizations are just playing musical chairs with executives right now because they think people are the problem, the strategy is the problem, and what we’ve been trying to educate the market on—and our customers are definitely seeing the benefits of—is the really tight collaboration that’s required between IT and the business.
It’s got to be a really collaborative effort, not a tennis game of a ball lobbed back and forth. You’ve got to do it together. So many organizations have not had that experience of being able to collaborate across functions, across teams. But that is the big breakthrough here is you’re able to really bust through the silos of systems, of teams, of data, to create very business objective-driven types of products and experiences.
You’ve managed to make models that cost as little as $700,000 to train. Is that cost-efficiency primarily due to embracing synthetic data?
A model is just really three components: algorithms, data and compute, and you can improve the model by scaling all three things. The real cost advantage for us comes from algorithmic improvements as well as synthetic data.
On the synthetic data front, synthetic data really needs a rebrand, because it sounds like kind of bad, but really it is a precision data set pre-wired to be consumed by the algorithms, and we synthesize it to look like the data that the model is going to most benefit from having in its training data set. A huge benefit, of course, is that these become IP friendly and commercially safe models.
You relaunched your former startup as Writer in 2020 and have evolved its mission over the years. What are the pros and cons of having such an adaptable approach?
The pro here is we can really react and respond to the capabilities of the foundational models that we are building and the foundational technology. The con is, it’s hard to just be reinventing so much all the time, but it’s what you’ve got to do to remain relevant.
It really is just a constant race to lead the customer. I think a big part of our competitive advantage here is that we build our own models, and that’s important because we’re able to anticipate what’s coming next and productize in advance of that.
(Except for the headline, this story has not been edited by PostX News and is published from a syndicated feed.)
