Who will nail the UI for AI?

Welcome to Runtime! Today: Why ServiceNow's new launch points to a new set of priorities for enterprise software design, CoreWeave finds another $14 billion, and the latest funding rounds in enterprise tech.

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

You can't click a clanker

The rise of cloud and mobile computing 15 years ago ushered in a new wave of design thinking in enterprise software, as hundreds of startups flooded the market armed with the idea that business software should be as easy to use as consumer software. That strategy worked, but if generative AI turns out to be a similar seismic shift in enterprise software development, it will force companies to throw out many of those design principles based around the mouse, browser, and mobile touchscreen.

"Why do I have to be stuck in what is called the WIMP interface — windows, icons, menus, and pull downs — that was invented in Xerox Park, right, 50 years ago?" former Google CEO Eric Schmidt wondered earlier this year, and on Tuesday ServiceNow unveiled its take on what an enterprise software user interface based around generative AI might look like. It introduced AI Experience, a collection of AI agents that respond to voice and text prompts within ServiceNow to find information across a company's environment.

• For example, Web Agents can help users set up a new computer — one of the most common tasks in the ServiceNow universe — by browsing the web for the user to find links and extensions for the external tools they need.
• AI Lens can extract data from static images or screengrabs to execute tasks based on that data.
• The agents will still be part of the existing ServiceNow workflow, because nobody wants to throw out non-AI or "deterministic" workflows that take predetermined actions in response to new events, said Amy Lokey, executive vice president and chief experience officer, in a press briefing.
• "What we're introducing now are non-deterministic workflows through AI agents that we're training to do particular tasks, to work with other systems, to read and write into those systems, do background work like research, [and use] web agents that can write to other applications," she said.

In the rush to show the board of directors that they have an AI strategy, lots of companies are slapping chatbots and agents onto tools designed before the launch of ChatGPT and calling it a day. As Mastercard's George Maddaloni shared last week, AI tools probably won't have much of an impact on employee productivity unless they are designed to connect to the existing sources of data employees need to do their jobs, and right now it takes some work to blend the old and new.

• "Decades of innovations have only reinforced silos and added complexity, slowing our game by making employees conform to the way software works — rather than providing solutions that work the way humans do," ServiceNow's Amit Zavery, president, chief operating officer, and chief product officer, said in a blog post.
• Like every other enterprise software company on the planet, ServiceNow is trying to convince customers to put its software at the heart of their AI strategy.
• Tools like AI Experience were designed in hopes of making it easier for ServiceNow users — IT departments, finance teams, and customer-service agents — to do their jobs by forging those connections to internal and external data sources within the context of their familiar ServiceNow experience.

Right now, however, no one company can claim to have solved the tricky integration of user experience and agentic AI performance, and there are lots of enterprise software users who would argue most vendors never got the pre-ChatGPT experience totally right either. Earlier this month Workday announced it had acquired Sana, a slick design-focused AI agent building startup, on the same day it introduced a very similar tool.

• True platform shifts tend to produce winners built from the ground-up right as that shift is taking place, while companies that rose to prominence in a very different era tend to have difficulty translating that success into the new era.
• Companies like ServiceNow are betting they can figure out the new UI archetype before their customers move the data needed to make agentic AI work to another platform.

They're not for the metaverse

CoreWeave won another substantial AI infrastructure contract this week, signing a $14.2 billion deal with Meta for GPU capacity through the end of 2031. CEO Michael Intrator told Bloomberg that the deal "is clearly a step in the right direction for diversification,” which was a concern for potential investors after CoreWeave's March S-1 report revealed it depended heavily on Microsoft for revenue.

It's also a good deal for Nvidia, which signed a separate $6 billion deal earlier this month to purchase unsold CoreWeave capacity over the next several years. CoreWeave still relies on a small number of companies for the vast majority of its revenue, but Microsoft, Nvidia, OpenAI, and now Meta have all pledged to spend billions on its infrastructure over the next several years.

According to Bokeh Capital Partners chief investment officer Kimberly Forrest, the deal signals "limitless" demand for AI chips, which feels like the kind of statement that is almost guaranteed to backfire. Still, it does look like Nvidia will be able to sell as many of its Blackwell generation chips as it is able to make over the next few years, although Bloomberg also reported Tuesday that Meta just acquired a chip startup to further work on its own AI designs.

Enterprise funding

Nscale raised $1.1 billion (!) in Series B funding (!!) for its AI neocloud ambitions, which it said was the largest Series B round in European history.

Cerebras scored $1.1 billion in Series G funding as it builds out its novel AI processor designs and infrastructure services.

Vercel scored $300 million in Series F funding, which values the web application development platform company at $9.3 billion.

Modular raised $250 million in Series C funding as it develops an AI development platform for building applications across different types of GPUs and CPUs.

Modal Labs scored $87 million in Series B funding for its serverless AI inferencing service.

Factory AI landed $50 million in Series B funding as it continues to develop agents for software development.

The Runtime roundup

Crusoe announced that the first phase of the Project Stargate data-center buildout is live in Abilene, Texas, under the operation of Oracle's cloud unit.

Asahi was forced to stop production after it was hit with a cyberattack Monday, delaying shipments of Runtime's favorite Japanese beer around the world.

Thanks for reading — see you Thursday!