Cursor chases designers; Glean's enterprise robots

Welcome to Runtime! Today on the last Product Saturday of 2025: Cursor wants to give designers the same AI tools as developers, Glean's agents get autonomous, and the quote of the week.

(Please forward this email to a friend or colleague! And if it was forwarded to you, sign up here to get Runtime each week.)

Ship it

Seeing is believing: Cursor's tools are widely used across forward-thinking software development organizations around the world, and just about all of those teams rely upon designers to shape their code into something useful and attractive to the end user. This week Cursor introduced a visual editor for its Browser tool that allows designers to tweak elements of a web application and automatically generates the code needed to enforce those changes.

The tool also lets users select elements on a page and describe how they'd like to change them with natural-language commands, like "make this bigger," the company said in a blog post. Sometimes ideas can get lost in translation when designers and developers try to collaborate from different points of view, and Cursor thinks its visual editor "kind of melded the design world and the coding world together into one interface with one AI agent," head of design Ryo Lu told Wired.

French connection: Mistral tends to get overlooked in discussions about the progress of frontier AI models, but the Parisian startup more than holds its own against its bigger American rivals. This week it introduced Devstral 2, a new coding model that the company said was "up to 7x more cost-efficient than [Anthropic's] Claude Sonnet at real-world tasks."

Devstral 2 is an open-weight model, but the company said it was "released under a modified MIT license," while a smaller version of the model was released under the permissive Apache 2.0 license. Mistral also introduced a command-line interface for the model called Mistral Vibe, which "explores, modifies, and executes changes across your codebase using natural language — in your terminal or integrated into your preferred IDE via the Agent Communication Protocol," the company said.

Agent, take the wheel: Once businesses get comfortable with putting AI agents into important parts of their business workflows — which is taking longer than once hoped — enterprise software companies are betting they'll want to turn it up a notch and embrace "set and forget" autonomous agents. Glean has already released several agents that can find nuggets of information across a company's data stores, and this week it previewed new autonomous capabilities for those agents.

"Glean’s new autonomous agents interpret instructions, make decisions, adapt as they go, and communicate their reasoning, eliminating the need for rigid workflows or low-level scripting," the company said in a press release. The autonomous agents are in beta for now, but the company also announced a new connector between its core search product and Microsoft Dynamics 365 that is generally available.

Roll 'em up: Autonomous or not, if companies start deploying AI agents at scale they're about to discover that older tools and techniques built for securing corporate applications weren't designed for that kind of world. This week Veza introduced a new service that promises to help customers secure agents provided by third-party companies or built internally.

AI Agent Security helps companies inventory agents running across their networks, understand how those agents interact with corporate data, and set governance and compliance rules. "In the end, all roads lead to identity: understanding who or what has access, and why," Veza co-founder and CEO Tarun Thakur said in a press release.

A graph for graphs: Neo4j's graph databases were the secret sauce behind Klarna's decision to consolidate data from its different SaaS applications into a single data model, which was a hot topic of discussion earlier this year. But larger Neo4j customers often find themselves running several different graph databases across different teams or locations, and this week the company introduced a tool for keeping track of that activity.

Neo4j Fleet Manager is "the first unified control plane for managing and monitoring graph databases across any environment: cloud, hybrid, and on-premises," the company said in a press release. As with Veza's tool, it is amazing that midway through the decade, simply understanding the scope of applications and databases running inside a company's environment remains an enormous problem.

Looking for a way to support independent tech journalism this holiday season? For $10 a month, you'll help us continue our mission to bring reliable and actionable coverage of this vital sector of the economy and gain access to supporter-only features currently in the works, such as an exclusive discussion and events forum.

Stat of the week

Menlo Ventures had some good news this week for beleaguered enterprise software vendors desperately hawking agentic AI products and services as the year comes to a close. In 2024, enterprise AI spending was split roughly in half between internal projects and external services, but at this point in 2025, "76% of AI use cases are purchased rather than built internally," according to its survey of enterprise AI buyers.

Quote of the week

"We are orders and orders of magnitude better for energy and cost." — AWS's Byron Cook, vice president and distinguished scientist, describing how neurosymbolic AI could solve generative AI's nagging hallucination problem without boiling the oceans.

The Runtime roundup

Broadcom CEO Hock Tan confirmed that earlier this year Anthropic ordered $10 billion worth of Google's Ironwood TPUs, and said the model provider recently put in another $11 billion order for the custom chips made in part by Broadcom.

Home Depot's cloud infrastructure was open to just about anybody for a year after an employee uploaded a private access token to GitHub, according to TechCrunch.

Thanks for reading — see you Tuesday!