OpenAI’s Dev Day: Is the bot revolution finally coming?

Welcome to AI Decoded, Fast Company’s weekly LinkedIn newsletter that breaks down the most important news in the world of AI. If a friend or colleague shared this newsletter with you, you can sign up to receive it every week here.

OpenAI’s GPTs: Is the bot revolution finally happening?

When I interviewed OpenAI COO Brad Lightcap last month, I asked him how his company is thinking about evolving its large language models (LLMs) and ChatGPT. Among other things, he told me ChatGPT would evolve beyond its current capabilities (namely, regurgitating text and summarizing documents) to carry out more complex tasks on the user’s behalf. That vision seems more plausible after seeing OpenAI’s developer event on Monday, where the AI darling announced new tools provided to both consumer ChatGPT users and developers for building more versatile AI-powered apps.

Speaking to an audience of 900 developers Monday, CEO Sam Altman announced a new, customizable form of ChatGPT that OpenAI has branded as “GPTs.” “You can build a GPT for almost anything, with instructions, expanded knowledge and actions,” he said. And because these agents run on large language models, you don’t need coding skills to build them. This gives the ability to create AI agents to far more people than just developers and coders.

Gartner analyst Arun Chandrasekaran tells me the biggest beneficiaries may be people within big companies or institutions who want to automate the work of their departments. For example, he says, a marketing department executive with no programming-language knowledge might train an agent on all the department’s style and brand guidelines and practices—without relying on IT staff for permission or help in doing it.

Altman used the GPT Builder tool to create a GPT chatbot during his keynote—an agent called Startup Mentor that can give advice to startup founders. To train the agent, Altman uploaded two papers he’d written containing helpful tips for founders. Had he wanted to, Altman could have given the agent a “custom action” (formerly called “Plug-ins”) that reaches out to a third-party source of information via an API.

While watching Sam Altman use the company’s new GPT Builder, and after hearing that OpenAI will soon showcase popular GPTs in a “GPT store,” I was reminded of the 2008 launch of another app store: Apple’s App Store. Apps gave us a totally new user interface for visually organizing our digital stuff on mobile computers, and then on all our devices. But apps aren’t necessarily forever. They’re not a perfect UX—we still have to stare at a screen and then tap and tap until we find what we want. Since at least 2016, tech leaders (like Microsoft’s Satya Nadella) have been predicting that bots will replace apps as the key motif for digital stuff. Bots, after all, enable a more hands-free, conversational way to access and create data. Obviously, bots have yet to take over the world, but a lot has happened in AI since 2016. Now we have “agents” backed by powerful large language models that are quickly gaining new skills. I met several developers at Monday’s event who believe that the agents OpenAI is enabling will soon become the dominant tool for mediating between humans and digital content. At some point, the narrative goes, they may even replace the concept of apps altogether. For now, it’s most likely that agents will become an important feature within apps—a way of interacting with apps.

Why AI-powered phones and smart speakers aren’t far off

Even if a bot revolution happens, AI agents still need to run on some kind of consumer hardware device. And Google and Apple, not OpenAI, own the operating systems of our current go-to devices: smartphones. Like Meta, OpenAI may not be content operating an app on someone else’s hardware platform. Indeed, Altman hinted at building an OpenAI device. “I do believe that every time a new . . . technology of this magnitude comes along, there’s supposed to be an amazing new computing device,” Altman told reporters on Monday. And he’s not denied reports that his company is working with ex-Apple design guru Jony Ive on some kind of AI hardware device.

Now that OpenAI is working to let developers (and end users) train agents on personal and third-party data, the possibility of a far smarter Siri-style assistant becomes clearer. Right now, the best way to interact with such an assistant may be via an app on our phones. But it’s not ideal. You have to wake up your phone and click a few times to get to the assistant, and it can take extensive prompts to bring up a useful response. Since LLM-powered agents are a conversational way of computing (you talk them through tasks and problems), wouldn’t it be better to have a hardware device that’s always there and always listening? Before you raise privacy objections, remember that we already have such devices: the smart home speakers Echo and Alexa. Such devices, which sit in your living room or kitchen and are always listening, seem like a natural vehicle for an AI agent.

“You still need something when you’re walking around the house,” U.K.-based hardware and software developer Spencer Bentley told me Monday. (He also happens to be the moderator of OpenAI’s developer forum.) “It’s got your entire routine, it’s got your email, your entire work history, every movie you’ve watched, every argument you’ve had, every bad thing you’ve posted online,” he says. “And it understands your emotions—if you’ve had a bad day [it might say] ‘I’m ordering extra beer for the refrigerator.’”

I asked Bentley why he thinks big tech companies like Apple and Amazon have, so far, not produced such a home AI device. “They’ve got people in their own companies shouting at them all the time, saying ‘we’re losing!’ . . . I know this because I talk to them,” he says. Such companies have the money and people to produce products quickly. But, Bentley says, the leadership fails to comprehend how fast AI agent technology is advancing (he believes the conversion to AI agents will take place in less than 10 years). “It took us 100 years to get from the first mechanical vehicle to get to the moon,” he said. “AI is moving much, much faster.”

Why OpenAI must pay for copyright claims against its users

OpenAI’s newly announced Copyright Shield is a pledge to indemnify developers against legal fees from copyright infringement lawsuits related to content generated by OpenAI’s models. OpenAI already made the same pledge to users of its ChatGPT Enterprise product. Why is it extending the indemnification to independent developers now?

It’s about investment. Thousands of small-scale developers are at this moment using OpenAI models and tools, usually through an API, to invent new products, including agents. Paying for that API access isn’t cheap, and when the developers’ app starts getting customers it gets even more expensive. So, they need money. And investors will not be thrilled about sinking money into a business that could immediately be sunk by a copyright infringement claim. OpenAI itself is defending against four separate suits from content owners claiming that the company’s AI models violated their copyrights. OpenAI can survive that. A fledgling startup probably can’t.

It’s just one of many things OpenAI is doing to attract developers to its ecosystem and keep them there. It makes sense at a more basic level. The developer had nothing to do with how the OpenAI models were trained, and OpenAI owns the models that are generating the potentially copyright-infringing content and sending it out through the API. Adobe also pledges to indemnify users for any legal issues arising from use of its Firefly image-generation tool.