Two years ago, OpenAI seized leadership in the AI race by introducing ChatGPT, a chatbot seemingly capable of all sorts of tech wizardry. Its simple, user-friendly chat window was a viral hit, forcing rivals to play catch up to avoid their hugely profitable businesses becoming obsolete.
Now, after tens of billions of dollars in spending, those rivals—Google parent Alphabet and Facebook parent Meta, among others—have closed the gap. And newcomer DeepSeek, a China-based company that wasn’t even on the radar a couple months ago, is nipping at the heels of the leaders after last month releasing a free, open-source model that’s largely as good as OpenAI’s, but at a fraction of the price.
Suddenly, OpenAI is on the defensive. As increased competition and cheaper open-source alternatives make AI models increasingly commoditized, can OpenAI keep its momentum?
Many industry experts say OpenAI’s focus on building ever-larger, more capable AI models no longer guarantees its future. Instead, to gain and keep users, it must offer better ways for customers to interact with its AI through even more intuitive, unique, interactive product design and tools.
“While the underlying models represent some of the most advanced computational feats of our era, our interactions with them remain rudimentary, as if we’re merely issuing commands, or ‘prompts,’ to a glorified machine,” Joanna Peña-Bickley, a former product design leader at Uber and Amazon and co-founder of Vibes AI, told Fortune.
For OpenAI, the stakes are huge. It’s pursuing a fresh funding round that would double its valuation to an eye-popping $300 billion—four times that of tech industry giant Dell and more than nine times the value of eBay.
Meanwhile, Elon Musk’s $97 billion bid this week to gain control over OpenAI’s nonprofit assets could complicate the future of OpenAI’s for-profit operations. OpenAI CEO Sam Altman has said OpenAI isn’t for sale, but Musk’s unsolicited offer could be a distraction that slows product innovation.
Losing momentum would put some of OpenAI’s market share in jeopardy and the huge amount of revenue that goes with it. With demand for AI products growing so quickly, any small setback could translate into billions of dollars in lost deals.
In this competitive arena, the winners will be those that both innovate their underlying technology while also adding cutting-edge design into their products that can be patented—and therefore can’t be easily copied by others, Peña-Bickley explained. It’s not just about aesthetics, she said, but building a competitive moat.
One simple example, she said, would be adding AI-powered voice recognition to ChatGPT logins that could also, in turn, detect AI voice deepfakes trying to hack into accounts. Another valuable feature could be a version of ChatGPT’s voice models that work on multiple devices without “latency,” or sound delays. That is, not just voice-enabled ChatGPT on the phone and desktop, but on other hardware too like earbuds, glasses, or home devices.
Amazon, through its $8 billion investment in Anthropic, she pointed out, is already moving in this direction. Anthropic’s Claude models will be used as the foundation of Amazon’s revamped Alexa device, as Reuters reported in August.
On Wednesday, Altman acknowledged that OpenAI has work to do when it comes to improving its various tools. In a post on X, he wrote that he wants to do a “much better job simplifying our product offerings.”
Altman pointed, in particular, to a drop-down bar on ChatGPT that gives paid users a confusing choice of models to tap for their query. There’s GPT-4o, o1, o3-mini, o3-mini-high, and more. The names aren’t exactly accessible to the masses. Instead, he wants AI to “just work.”
ChatGPT opened the door to the AI boom
According to OpenAI research scientist Ariel Herbert-Voss, OpenAI created a game-changing chat box interface when it introduced ChatGPT in November 2022. Now, it must invest more to help people discover new, creative ways to use AI while also understanding its limits.
Before ChatGPT, he explained, the large language models available were difficult to use because they didn’t have chat windows. Users needed technical knowledge to access the technology, severely limiting its appeal. “You had to build your own windows, basically everything else from scratch,” said Herbert-Voss, who is now founder and CEO of AI security startup RunSybil. “A lot of non-experts churned off because it wasn’t clear what they could use it for.”
While the simple chat box ultimately used turned out to be revolutionary, and ChatGPT usage is booming, it’s no longer enough to stay ahead when every other LLM has one too. Rather than just a question and answer conversation with a chatbot, what’s needed now are more interactive, collaborative and creative tools that let users explore what generative AI can do.
This already exists in some other software. For example, Adobe’s Firefly generative AI tool lets users create and edit images directly within Photoshop. Meanwhile, Microsoft Copilot, which uses OpenAI models, is embedded in Word, Excel, and Outlook to help users draft emails, summarize meetings, and automate spreadsheets.
Rowan Curran, a senior analyst at market research firm Forrester, noted that OpenAI’s product challenges are not simply related to design. The company also lacks a comprehensive suite of tools, infrastructure, and services that go beyond just having a powerful AI model. Big Tech companies with cloud computing arms like Amazon, Google, and Microsoft offer entire platforms that businesses can use to easily deploy AI models, customize models with their own data, and integrate AI into their workflows.
Instead, OpenAI seems focused on improving the performance of its models as a way to attract business customers, Curran explained. That’s concerning, he said, given that OpenAI’s leadership in models versus open source competition like DeepSeek is shrinking—particularly in cases when using the highest-quality models isn’t necessary.
OpenAI has already made some innovative product moves
Not everyone agrees that OpenAI is lacking on the product front. Aaron Levie, CEO of cloud content management company Box, argues that OpenAI has already tilted its efforts toward creating software on top of its models, and moved away from focusing on just the models.
He called the company’s Operator, an AI agent released in January that is designed to autonomously perform tasks on the web for users, “a huge breakthrough.” He also pointed to OpenAI’s recently released Deep Research, which can autonomously handle in-depth, multi-step research tasks online. For now, both are only available as “research previews” to customers on OpenAI’s $200 per month Pro plan.
“I think we’re seeing a great, incredible rate of innovation from this team, it’s going to be very competitive,” Levie said, “though nobody can afford to rest on their laurels.”
Improving OpenAI’s products goes beyond changing their design. Tweaking the underlying models to make them better at reasoning, memory, and processing text, audio, and video can also do the trick. Initially, for example, OpenAI was unable to search the internet in real-time, and therefore couldn’t always provide up-to-date responses. But now it can, without any change to the design that users see when they enter queries, said Curran.
Another departure by OpenAI is upending how people interact with its tools. A critical example is OpenAI’s work with AI agents such as Operator, which, after getting initial directions by users, acts autonomously on their behalf.
Instead of clicking through menus or writing code, people can simply chat with OpenAI’s agents—similar to communicating with a colleague on workplace messaging service Slack. “The next evolution of AI isn’t about building interfaces,” AI consultant Reuven CohenCohen said. “It’s about interacting with autonomous agents that operate within our digital workspaces, transforming how we engage with complex systems.”
Could great design make OpenAI profitable?
Still, Peña-Bickley said that while OpenAI has already set a high bar with ChatGPT, Deep Research, and Operator, great design could make OpenAI profitable by making users more willing to pay for its tools and use them more. In September 2024, OpenAI said it expected nearly $5 billion in losses on $3.7 billion in revenue for the year.
“Imagine an interface that doesn’t simply process requests behind a static pane of glass, but instead reveals its inner workings in a way that builds trust rather than adding complexity,” she said.
For example, the AI could use “gentle sound design” to make the AI feel more natural and intuitive, Peña-Bickley suggested. Instead of just showing a spinning wheel or a loading bar, it could use subtle sounds to signal what it’s doing—like a soft chime while gathering information, a deeper hum when analyzing complex data, and a satisfying tone when it’s ready with an answer. Such little design touches would make interactions feel smoother, clearer, and more human-like.
OpenAI has made some important hires that make it clear that it recognizes the importance of product design, Peña-Bickley said. In addition to hiring its first chief product officer, Kevin Weil, in June, OpenAI has also hired in an unknown capacity Jony Ive—who once led industrial design at Apple for products like the iPhone, Apple Watch, and Mac—as well as two key people from his former Apple team, Tang Tan and Evans Hankey. In November, OpenAI also hired former Meta hardware lead Caitlin Kalinowski to lead the company’s robotics and consumer hardware efforts.
Altman recently addressed OpenAI’s growing product efforts last month in a personal blog post reflecting on the past two years since the release of ChatGPT: “Our vision won’t change; our tactics will continue to evolve,” he wrote. “For example, when we started we had no idea we would have to build a product company; we thought we were just going to do great research.”
Finding the right balance between research and products
But Herbert-Voss cautions that while OpenAI’s culture is evolving to be more product-focused, at its heart it is a research company. “The researchers are kind of the first-class citizens, and then everything else kind of flows out from that,” he said.
It may be tricky to find the right balance, he added. “When I was at OpenAI, the thinking was if you don’t have a good model, then everything else kind of doesn’t matter,” he said. “But if you’re trying to build a product company versus a research program, you are going to have to invest a lot more resources into product.”
Box’s Levie, however, maintains that OpenAI’s ChatGPT remains the ubiquitous “Kleenex” of the AI space—and that the company has little to worry about when it comes to the success of its model and its products. That is, as long as their offerings remain state-of-the-art. “They can’t ever be 5% or 10% behind,” he said. “But they don’t need to be 50% ahead. They have to be just ahead enough where if you go there you’re not losing any sort of value by doing your query in ChatGPT.”
