AI startup writers have a client roster like Uber, Salesforce, Inuit, and Qualcomm, just like Fortune 500 characters. The San Francisco-based company offers a suite of AI tools designed to deploy agents and integrate the technology into business workflows.
The writer was co-founded in 2020 by May Habib, who previously launched a startup called Qordoba in 2015, a company focused on natural learning processing (NLP). Although she has been relatively close to the hub for generating AI recently, she has crossed fundamental technologies in a decade. “I've been doing this for 15 years,” she told Observer in a May interview.
The writer's “end-to-end agent builder platform” is powered by its proprietary Palmyra model family. Unlike competitors with high training costs, writers touted their cost-effectiveness—some of the newer models have just $700,000 in training.
This approach not only attracts trademark customers, but also attracts significant investor interest. In November 2024, the writer raised $200 million in the C Series season, with an estimated $1.9 billion in the company. As CEO, Habib reportedly owns a 15% stake, estimated at $285 million.
She thinks the company is just beginning. “It feels like Agent AI is going to eat a lot of people's labor,” she said. She added that it's hard not to be excited about “watching people automate what happened in a lot of things they've done before,” she added.
Observers caught up with Habib to discuss the startup’s role in the field where AI agents are constantly competing. The following conversation has been edited for length and clarity.
Observer: When did you realize and become interested in AI?
May Habib: I think people in AI are two different species – pre-chatgpt species and behind-the-body species.
We work in NLP when we build an aligned dataset for statistical machine translation models. So even true deep learning techniques. Even though we came up with the A Series for the writers, we talked about machine learning, and talked about NLP and talked about Transformers, we never used the term “AI” because it was a bit taboo.
You have a wide range of customers. How much do they pay for these tools, and what are your favorite use case examples?
We sell to Fortune 500 companies – these people spend $100 million a year. Writers have about $2 million or $3 million, which is the highest ROI part of their investment. We know they need to try everything, and what we really try to do crazy continues to prove that these things work on a large scale.
Use cases vary greatly depending on vertical industries and industries. In Payer Space Services, like your health insurance company, some of our favorite use cases in production are related to helping members really make the most of their plans.
In medicine, some of the best use cases are helping salespeople in the field better understand who they are selling to and ready for a conversation, which can take hours of preparation as you are reading such intensive research material.
exist [consumer packaged goods] Space, watching in real-time people optimize their list of products on Amazon.com and Walmart.com is really exciting to be able to really sell more.
In retail, people are using writers and agents to combine emotions and customer feedback in a very, very specific way, and then act on that.
So just use tools like the author, which helps drive killer use cases that are really high-income growth. We helped launch the Airbnb experience – the new products they launched were launched with the writers. We wrote 37,000 pages and without the technology they wouldn't have done that at all.
What would you say to people who are worried that AI will take jobs or in some ways replace labor?
People refused to write as a technology thousands of years ago because it would make our memories weaker. This is just the standard of human curriculum. We will writh our hands on any way that changes people make a living or their way of living, and it will undoubtedly fundamentally change our way of living, but it will never be more exciting.
None of our 300 clients lay off employees here. Everyone has mountains of characters related to AI, and they can't find anyone to do it. I think we will be fully employed to make the most of this incredible productivity, but we have to make sure we are building a truly accessible experience and joining the market in a very fair way. But there is no doubt that this is an overwhelming positive thing for humans.
A recent report by the authors analyzed AI adoption in the workplace and found that two-thirds of leaders said this caused internal tension. What do you think is the correct way to use navigation?
I see this is the tension caused every day. So many organizations are just playing music chairs with executives now because they think people are the problem and strategies are the problem and we are always working to educate the market – the benefits our customers will certainly see – it’s really close collaboration between it and the business.
It must be a very cooperative effort, not a tennis game with back and forth. You have to do it together. As a result, many organizations have no experience in being able to work across teams and functions. But it's a big breakthrough, you're able to really destroy through silos of systems, teams, data to create a very business-driven type of product and experience.
You've managed to make models that cost as little as $700,000 in training. Is cost efficiency mainly due to the inclusion of comprehensive data?
A model is actually three components: algorithms, data, and calculations, and you can improve the model by scaling all three things. For us, the real cost advantage comes from algorithm improvements and synthetic data.
In terms of synthesizing data, synthesizing data does need to be rebranded because it sounds like it's a little bad, but in reality, it's a pre-connected exact dataset that can be consumed by the algorithm, and we synthesize it looks like the model gets the largest data in the training dataset. Of course, one huge benefit is that these benefits are IP-friendly and commercially secure models.
You relaunched your former startup as a writer in 2020 and have been growing its mission for years. What is the method to adopt this adaptability?
The expertise here is our ability to truly react and respond to the underlying models and underlying technologies we are building. The scam is, it's hard to reshape all the time, but it's what you need to do to keep it relevant.
It's really just a continuous competition to lead clients. I think a big part of the competitive advantage here is that we build our own models, which is important because we are able to anticipate what will happen next and produce before that.
