While both writing and coding modes provide the ability to request direct editing, the split user interface for canvas is designed with one additional set of shortcuts for those focused on AI-assisted writing and another for developers. In the demo, Levine showed how you can utilize author abbreviation to condense the word count on a canvas or attempt to “put the finishing touches” on a draft. He also used one of the more lighthearted shortcuts to add some random emojis. On the developer side, ChatGPT can add logs, comments, and attempt to resolve issues in the workspace.
ChatGPT saves different versions of the canvas as you iterate, so you can revert to elderly versions if you prefer this output. Authors who may be concerned that what they submit will be used by OpenAI to train the model should go to their user settings and make sure “model training” is turned off.
By allowing ChatGPT to make changes and suggestions, OpenAI blurs the line between authorship and word selection. As someone who works with professional editors on a daily basis, I’m skeptical that the canvas beta will match their insightful notes and precise directions. However, for people who don’t have basic access to human writing partners, I can see how it would be beneficial to get synthetic notes on the structure and content of a composition.
It’s worth noting that the three people listed in Project Canvas as “supportive leadership” are no longer with the company. Former post-training co-leader and co-founder John Schulman left in August and now works at Anthropic, a rival artificial intelligence company. Additionally, former chief technology officer Mira Murati and vice president of research Barret Zoph stepped down from their positions a week before the launch. At a press conference at OpenAI’s office following the departure of current chief product officer Kevin Weil, he reaffirmed the company’s commitment to continuing to release the software.
“I think 2025 will be the year that agent systems finally go mainstream,” he says. The idea of an AI “agent” that can not only perform software tasks with you, but is also agile enough to be sent out into the digital wilderness to do things on your behalf, is both the recent past and the anticipated future of generative artificial intelligence.
Last year, the WIRED project discussed ChatGPT plug-ins that people could utilize for tasks such as booking flights on Expedia or making reservations with OpenTable — arguably a step toward more “agentic” AI tools. However, the plugins were later discontinued and more narrow custom GPT chatbots were launched in their place.
With this in mind, the Canvas beta release appears to be another attempt to enhance AI models with greater decision-making capabilities, which could lead to surprises. During one WIRED demo, Levine highlighted part of the canvas and asked for an edit, and ChatGPT then made a change to the line at the bottom, beyond its highlight. “The really interesting thing is often that if you highlight a section, it will cause a change in that part,” he says. “But ChatGPT has the ability to decide where to edit.”
The closest alternative currently available to OpenAI’s canvas tool is probably the Google Gemini integration, which allows you to utilize generative AI in Docs or Anthropic artifact tool. Chatbots are definitely not dead, but AI companies are now realizing the limitations of the format and are looking for ways to diversify their software to discover novel sticky user interfaces. Google recently gained recognition in tech circles for its entertaining podcasts about artificial intelligence – even CEO Sam Altman praised the tool.
As billions of investment dollars continue to flow through Silicon Valley to AI companies, consumers can expect to see more structured experiments based on existing tools, such as AI podcast hosts and AI document editors, being released regularly over the next year. The chatbot race is far from over, and future versions of the technology will likely move away from the gray chatbox towards a more multi-faceted approach.