OpenAI’s Attempt to Catch Up With Anthropic Could Cannibalize Its Business
OpenAI closed out its big week of fundraising news with a rather understated product announcement: the release of Canvas, a new ChatGPT interface that makes it easier for users to draft documents and code.
That might not sound like a big deal. But it has bigger implications for the company’s position in the AI race and its relationship with its customers than you might think.
With Canvas, users can open up a separate window beside ChatGPT’s normal chat window to work on writing and coding projects. There, they can highlight specific sections of their text or code for OpenAI’s model to edit, or use shortcuts for common actions like fixing bugs in their code or lengthening or cutting a piece of writing.
That’s a departure from the traditional ChatGPT interface, which typically requires users to scroll up in their chat history to find earlier responses or completely regenerate entire documents or lines of code from scratch if they’re not happy with what ChatGPT produced.
If any of those features sound familiar, that’s probably because many already exist in products from other AI startups, such as coding assistant Cursor and OpenAI archnemesis Anthropic. In some ways, OpenAI’s Canvas feels like a direct response to Artifacts, a similar product Anthropic first announced in June. (That’s perhaps unsurprising, given that the research lead for Canvas, Karina Nguyen, joined OpenAI from Anthropic four months ago.)
And although Canvas does boast some capabilities Artifacts lacks (like the ability to automatically translate code between different programming languages), it largely feels like OpenAI’s attempt to fill in the gaps in its offerings, a theme that we saw at its Dev Day event last week as well. (And we’d expect OpenAI’s product to have a couple extra features given that they released it four months after Anthropic’s!)
Anthropic’s Artifacts still has a leg up on Canvas in some areas. For instance, users generating code for a simple website or video game with Claude can see a prototype of what their product will look like next to their code. They can also share those prototypes, or “artifacts,” with others. And, Anthropic’s Artifacts product has been generally available to all Claude users since August.
OpenAI’s direct assault on Anthropic’s Artifacts highlights how competition between AI firms has become a race to build the best user interfaces as much as a race to develop the most advanced models. Increasingly, startups are moving away from simplistic chatbots to interfaces that allow users to make small tweaks more easily or share their AI-generated work with others.
That shift has broad ramifications for the near term future of the sector. For one thing, by focusing so much on its consumer-facing ChatGPT product, OpenAI could cannibalize the businesses of its customers, such as GitHub Copilot, Cursor and the dozens of other startups using OpenAI’s conversational AI to build coding assistants.
That reinforces the inherent conflict between OpenAI’s consumer-facing product (which generates most of the company’s revenue) and its business of selling access to its models to developers via an application programming interface. It could also make it more difficult for OpenAI’s biggest partner, Microsoft, to sell its API to developers who might be afraid that OpenAI will someday encroach on their business.
OpenAI also might be getting itself into a market it’s not fully prepared for. Today, it’s convenient to draft emails or short memos in ChatGPT, but many users still prefer doing longer-form projects using traditional workplace tools such as Google Docs or Microsoft Word, which are sure to add similar AI-powered editing capabilities as well.
To compete with those types of businesses, OpenAI will need to add a ton of new features that workers at big companies demand, like collaboration capabilities, user permissions, joint editing and templates. For OpenAI, that might be biting off more than it can chew.