Runtime: Cursor gets Grokked; TypeScript rolls a seven

Welcome to Runtime! Today on Product Saturday: Cursor releases its first AI model jointly developed with SpaceXAI, Microsoft makes TypeScript ten times faster, and more.

Please forward this email to a friend or colleague! If it was forwarded to you, sign up here to get Runtime for free every week, or level up here.

Coding, and beyond: It's been less than a month since Cursor announced it was joining SpaceX (now known as SpaceXAI_FINAL VERSION), but the two companies started working together on a new AI model several months ago. That group project arrived this week in the form of Grok 4.5, which Cursor said was "our most intelligent model and the first we've built for more than software engineering," in a blog post.

While the new model is still pretty good at coding, when paired with the Grok Build AI assistant it "is capable of building complex Excel models that involve research from the web, multi-sheet formula use, and even leaves stickies or notes behind for future reference," SpaceXAI said in its own blog post. It's also considerably cheaper than Anthropic or OpenAI's flagship models, although it could be a tough sell in the enterprise given Grok's history.

Many servers are typing: For a long time Slack appeared to be one of Salesforce CEO Marc Benioff's biggest M&A flops, but growing adoption of enterprise AI has put the collaboration tool in a very interesting position. This week Salesforce rolled out a new version of Slackbot that can surface Salesforce and third-party data within Slack through natural-language queries.

Slackbot can now "reason across the entire Headless 360 platform and the tools your team already uses, all in one place, thanks to dedicated new MCP servers from Salesforce," the company said in a blog post. The so-called "headless" strategy now in favor across some enterprise software companies allows customers to assign AI agents to work with SaaS data, rather than logging into a web-based UI.

GRC's reputation is shifting, and GRC Engineering is at the forefront. Join Ayoub Fandi, GRC Engineer at Lovable, and Justin Pagano, Sr. Director of GRC Engineering at Vanta, to discuss Legacy GRC and what comes next. You’ll learn practical ways to start automating, with lessons from Lovable and Vanta.

Learn more

Follow the script: The TypeScript programming language allows software developers to build JavaScript apps with a little more specificity, and now it runs a lot faster. After announcing plans to rebuild the TypeScript compiler in Go last year, Microsoft delivered TypeScript 7.0 this week.

"TypeScript 7 brings native code speed, shared memory multithreading, and a number of new optimizations that typically yield speedups between 8x and 12x on full builds," the company said in a blog post. It also improves application reliability: "Our data insights have shown us that TypeScript 7.0’s new language server has actually reduced failing language server commands by over 80%, and reduced server crashes by over 60% compared to that of TypeScript 6.0," Microsoft said.

Frame it: Mainframes have been affectionately (sort of) referred to as "Big Iron" for decades, but IBM decided they're ready to get a little smaller. This week Big Blue announced that the new z17 and LinuxONE 5 mainframes will be available in a rack mount option, "marking the first time IBM is offering rack mount alongside single frame systems across its full Z and LinuxONE portfolio," the company said in a press release.

Most conventional servers in self-managed data centers and hyperscaler regions are rack mounted, which allows administrators to easily swap in new hardware as needed, but mainframes were traditionally offered as one big box. The new options give data-center operators more flexibility over how they use their real estate without having to ditch their mainframe applications, which run a surprisingly larger portion of the Fortune 500 than you might think.

Call the tune: As companies start to experiment with open AI models in hopes of reducing their enterprise AI spending, they often find they need to "fine tune" those models around their internal data to get the best results. As part of its strategy to layer software services over its data-center operations, Crusoe launched a new service that makes it easier for companies to tailor their models without having to configure a lot of hardware.

Crusoe Serverless Fine-Tuning allows companies to "launch a fine-tuning job in a few clicks — select a base model from a curated library of top-performing open-weight models, upload a custom data set, configure settings with pre-configured best practices, and submit; no dedicated reservation required," the company said in a press release. The neocloud also added a new option called Self-Serve Deployments that allows customers to run those models on Crusoe's GPUs with a single click.

Join peers following The Stack on LinkedIn

Microsoft warned admins this week that the next Patch Tuesday will be a little more hectic thanks to AI models, which are helping Microsoft discover and patch more vulnerabilities than ever before. That could be a problem for smaller companies, Huntress' Muhammad Yahya Patel told The Stack: "The blog implicitly assumes a Microsoft 365 enterprise customer as its audience. For Windows deployments that don't fit that profile, the practical guidance is thin."

Google released a new coding agent that was designed to help developers tackle complex tasks like optimizing their own algorithms, as compared to vibe coding a prototype for the customer meeting. AlphaEvolve has already helped Google improve the efficiency of its own infrastructure, and it runs on the company's TPU chips.

When it comes to financial services customers, AWS, Google, Microsoft, and Oracle are now considered "critical third parties" by the UK government, which will require them to work more closely with financial regulators. The goal is to improve the resiliency of cloud infrastructure services to avoid disruptions to the markets, and now we wait to see which cloud provider goes down first thanks to a feature designed by a committee of quants.

Transformers are the atomic unit of the generative AI revolution, but IBM researchers think they might have come up with a better way to design large-language models. Known as Continued Fraction Generative Networks, the proposed architecture is a “practical and conceptual shift, pointing to lighter-weight generative AI models that perform competitively, and in many cases even better” than today's frontier models.

We're also reading:

Paul Stack takes a long look at how engineering teams are evolving six months after the latest crop of AI models jump-started the AI coding craze, warning that teams still sitting on the sidelines are digging themselves a hole.

As the enterprise tech establishment frets about frontier model companies stealing their thunder, Arvind Narayanan and Akash Kapur argue they have little choice to avoid becoming commodity players but to tackle enterprise software.

Want to sponsor a newsletter and get in front of 30,000+ subscribers, including CTOs and CIOs wielding $150 billion in annual tech budget? Get in touch through our Partner page.