Can OpenAI keep developers in its orbit?

Welcome to Runtime! Today: Following its latest round of upheaval, OpenAI continues its campaign to win the hearts and minds of developers, the race to develop AI coding assistants is getting a little out of hand, and the latest funding rounds in enterprise tech.

(Was this email forwarded to you? Sign up here to get Runtime each week.)

Ok developer

It must have been a relief for OpenAI executives to talk about anything but its recent leadership turmoil during Tuesday's DevDay event, which was a stripped-down affair compared to last year's inaugural event. Instead, Tuesday's presentation focused on nuts-and-bolts tools pitched to software developers, which 25 years after Steve Ballmer's epic keynote speech remains probably the most important constituency that any would-be enterprise tech vendor needs to delight.

OpenAI released four new capabilities across its APIs during the event, adding features such as speech-to-speech communications and prompt caching that developers will be able to use in their own apps. The announcements reflect just how competitive the market for AI developers has become over the past year, as startups like Anthropic and big vendors like Microsoft and Google race to deliver the tools that will draw developers on their platforms.

• The most significant announcement was the Realtime API, which will allow developers to build apps that can respond to spoken-word input with AI-generated voice responses.
• This feature is modeled on ChatGPT's Advanced Voice Mode, which got the company in trouble with Scarlett Johansson last year, but uses a different set of voices.
• OpenAI demonstrated an app built with Realtime API and Twilio's voice-calling APIs that could (theoretically) order food from a catering service from an upcoming event as well as an AI assistant app that could plan an upcoming trip using only voice commands and responses.
• This feature is in public beta now, and Techcrunch noted that at the moment OpenAI didn't provide any sort of disclosure that the API allows an AI bot to communicate with an unsuspecting human, which got Google into a lot of trouble a few years ago.

The other three announcements focused on adding new ways to contain costs and improve performance for developers using OpenAI's existing models. If the generative AI revolution is ever going to achieve escape velocity, it's going to have to do more and cost less than the current generation technology allows.

• Prompt caching allows developers to more efficiently reuse components across their apps, which OpenAI said would reduce costs by 50%. (Anthropic already offers caching that reduced developer costs by up to 90%.)
• Developers can now use images to fine-tune OpenAI's flagship GPT-4o model, allowing them to generate more accurate output within visual search or object-recognition apps.
• And Model Distillation "lets developers easily use the outputs of frontier models like o1-preview and GPT-4o to fine-tune and improve the performance of more cost-efficient models like GPT-4o mini," OpenAI said.

Tuesday's event certainly lacked the buzz that accompanied last year's first-ever Demo Day, and CEO Sam Altman sat out the morning presentation and demos. But according to Maginative's Chris McKay, maybe that's a good thing: "While these announcements lacked the fanfare of previous events, they mark a shift in strategy for OpenAI—away from headline-grabbing new products, and toward creating a more efficient and sustainable ecosystem for developers," he wrote.

• As AI models start to look more and more like each other, companies like OpenAI will need to find ways to convince developers that they offer something more than just access to GPT-4, which they can also get through Microsoft.
• And as Ballmer knew, once developers have made an investment around a company's tools they're likely to stick around for quite some time rather than going through the process of rebuilding their apps.
• OpenAI is fighting a slightly uphill battle here, given how many developers are already using a major cloud provider to build non-AI applications and could be easily convinced to use their AI developer tools.
• But every new generation of technology has always produced a new winner born around that platform shift, and assuming it can put together a few stable years OpenAI has as good a chance as anyone of meeting AI developers where they want to be.

A MESSAGE FROM CHRONOSPHERE

Chronosphere has been named a Leader in the 2024 Gartner® Magic Quadrant™ for Observability Platforms based on its ability to execute and completeness of vision. Gain complimentary access to the report to find out why Chronosphere was recognized as a Leader, learn how to best evaluate observability platforms, and gain valuable insights into the entire observability space.

Get the Report.

IANAL, but…

It's hard to get a startup funded these days unless you're working on something AI-related, but the folks behind PearAI might just be a lasting symbol of the excesses of the past two years. Its launch over the weekend caused quite the commotion within AI, open-source, and entrepreneurial communities.

Techcrunch has a good summary of the debacle behind the launch of the coding assistant, which at first glance looked interesting as a new attempt to capitalize on one of the most promising applications for generative AI. However, PearAI immediately drew criticism as a very close clone of Continue.dev as well as Microsoft's Visual Studio Code, the most popular development environment in use today.

Both Continue.dev and VSCode are open-source projects, but PearAI's founders hit a new low (or high, if you like comedy) in the generative AI era by using ChatGPT to write a license for their product called the Pear Enterprise License, which closed the source code. After a pile-on the founders admitted their mistake and ditched the Pear Enterprise License, but add one more thing to the list of tasks that you probably shouldn't use ChatGPT to accomplish.

Enterprise funding

Observe raised $145 million in Series B funding and launched Project Voyager, a new AI observability tool focused on incident management.

Eon landed $127 million across three different funding rounds from the past year for its plan to build a "backup autopilot" for cloud customers

Supabase scored $80 million in Series C funding as it continues its work on a managed version of the PostgreSQL database.

DataPelago landed $47 million in new funding and launched its product, which allows companies to process data stored in structured or unstructured formats on the hardware of their choice using open-source data engines like Spark and Trino.

Qodo raised $40 million in Series A funding for its AI code-testing and bug-funding platform.

Resolve AI scored $35 million in seed funding and introduced its operations management tool, which, believe it or not, uses AI to detect and repair production issues.

The Runtime roundup

Salesforce suffered a global outage that affected several services throughout the day on Tuesday, and as of publication time it was rolling out an "emergency release" to affected locations.

AWS announced plans to shut down several services, including AWS App Mesh, part of a continuing break from its long tradition of supporting services indefinitely under new CEO Matt Garman.

A MESSAGE FROM CHRONOSPHERE

Evaluating observability platforms? Get complimentary access to the 2024 Gartner® Magic Quadrant™ for Observability Platforms to see this year's evaluation and how the Observability space is evolving.

Get the Report.

Thanks for reading — see you Thursday!

This newsletter was updated Wednesday 10/2 to correct the name of OpenAI's Advanced Voice Mode.