|
Hi! If you’re finding value in our Applied AI newsletter, I encourage you to consider subscribing to The Information. It contains exclusive reporting on the most important stories in tech, like this story on how Amazon is reshaping the TV ads market. Save up to $250 on your first year of access. Welcome back! OpenAI has a long habit of launching products and features that compete with other tech startups that sell apps or services using OpenAI’s artificial intelligence. Now, OpenAI has its eye on the market for AI consulting services that help large enterprises develop their own AI apps—if they’re willing to spend at least $10 million, as Sri and I reported over the weekend. Oftentimes these apps are versions of ChatGPT, customized to the enterprises’ data. But OpenAI has ambitions to figure out ways to customize its models to the point where a company might be willing to spend $1 billion for them, presumably because they would help the company make or save billions of dollars! Perhaps OpenAI noticed the rise of Palantir and how consulting firms like PwC, Accenture and BCG have been flush with new revenue from large enterprises grappling with large language models. Data labeling firms like Scale AI and Turing also have gotten into the consulting game, as have a raft of AI startups that initially tried selling AI apps but pivoted to consulting services when they found the apps market to be too hard. Even ex-OpenAI CTO Mira Murati’s startup Thinking Machines has considered developing a custom-model business! OpenAI’s move into consulting-like service might be particularly awkward for PwC and BCG, which previously announced partnerships with OpenAI to advise customers on how to use OpenAI’s chatbots in the workplace. Spokespeople for those firms did not immediately provide comment. OpenAI thinks its edge is offering the services of in-house researchers who helped develop the startup’s foundation models and are now available to high-paying customers. That pitch could prove especially compelling with AI talent in such short supply, as evidenced by the ongoing melee between OpenAI and other firms over hiring and retaining such researchers. Here’s what else is going on… AI Data Wars Erupt A war has begun over who should be able to access customer data in enterprise applications like Slack, Notion and Atlassian’s Jira and Confluence, pitting the owners of those apps against AI startups like Glean and, potentially, OpenAI. The startups are using AI to help corporate employees to quickly find conversations, emails and other data stored across the many different applications and databases large firms typically use. While enterprise search has been around for decades, AI models can grasp the context of work interactions and return results quickly, making employees more productive. A few weeks ago we covered one of the early salvos, as Slack changed its terms of service to prevent API users from exporting and storing data, potentially hurting Glean. Now we’ve found that Atlassian has been throttling the APIs for Jira and Confluence—its two most popular products. Meanwhile, contact center automation firms like Genesys and Nice have started charging customers extra to process their audio data through third parties. Customers want to use Glean or OpenAI to search for and access their work files and data, no matter which app holds the data. But large software firms are threatened by those services because they let customers handle more of their complex computing tasks through third parties, potentially relegating them to database providers. As a result, they are reluctant to let AI firms have free access to the apps data, said Wade Foster, CEO of Zapier, which builds tools that let applications share data and handle tasks like auto-responding to website forms and adding emails to customer management systems. “I don’t think the industry has a great answer to how to handle [products like enterprise search], especially incumbents who see these AI apps as a threat to their business,” Foster said. The situation is reminiscent of changes that X and Reddit made in 2023, when they began charging for access to applications programming interfaces that they had previously made available to developers for free. In those cases, the companies realized they should be getting paid. In Reddit’s case, startups like OpenAI and Anthropic were using the data to train AI models, and OpenAI eventually struck a deal to pay for that access, and Reddit is suing Anthropic for allegedly training on its content without permission. There are valid reasons for software firms to set limits on how much other firms can use their APIs, ranging from a need to balance performance across all customers to fending off cyberattacks. But at a time when software providers are using customer data to train AI models and build new products, rate limits are also poised to become a competitive tool. In other words, the API data wars have just begun. –Kevin McLaughlin
| 
