Today on Product Saturday: former OpenAI CTO Mira Murati's new company introduces its first product, DeepSeek drops another low-cost model on the market, and the quote of the week.
Today: The latest in the long-running saga of enterprise tech marketing departments trying and failing to look cool, Oracle customers are receiving extortion attempts after a breach, and the latest enterprise moves.
For companies with sensitive data requirements or substantial investments in data centers, a hybrid cloud strategy offers the best of both worlds — and a significant management challenge. Eight members of our Roundtable discussed how to balance cloud and on-premises deployments.
Today on Product Saturday: former OpenAI CTO Mira Murati's new company introduces its first product, DeepSeek drops another low-cost model on the market, and the quote of the week.
Welcome to Runtime! Today on Product Saturday: former OpenAI CTO Mira Murati's new company introduces its first product, DeepSeek drops another low-cost model on the market, 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.)
Tinker tuner: Amid a wave of high-profile departures from OpenAI last year, Mira Murati's decision to leave and start her own company really stood out. Thinking Machines raised $2 billion on the strength of her reputation and the frenzy associated with any tech boom, and this week the company finally revealed what it has been working on over the last year.
"Tinker is a managed service that runs on our internal clusters and training infrastructure," and it "lets you fine-tune a range of large and small open-weight models, including large mixture-of-experts models such as Qwen-235B-A22B," the company said in a blog post. The process of fine-tuning allows companies to take off-the-rack large-language models and train them on their existing data to accomplish any number of tasks specific to their business, but it can be expensive and time-consuming to do on your own.
I, Claude(ius): The fierce competition between OpenAI and Anthropic to build a better coding model is really paying off for software developers, as it seems like each company is raising the bar on a monthly basis. This week Anthropic released several new features for Claude Code, including an extension for Microsoft's Visual Studio Code and an improved user interface for terminal jockeys.
The company also introduced "checkpoints," a feature stolen (complimentary) from video-game design that "automatically saves your code state before each change, and you can instantly rewind to previous versions by tapping Esc twice or using the /rewind command," Anthropic said in a blog post. The company said checkpoints allow developers to move more confidently when employing autonomous coding agents, which are still a little too cutting-edge for most enterprises.
Everyday low prices: DeepSeek sent shockwaves through enterprise tech in January when it dropped a very efficient large-language model that cost relatively little to train, although it became clear pretty quickly that the reaction was overblown. Still, there is definitely enterprise interest in cheap but capable LLMs, and this week DeepSeek released a new model with a feature that could help reduce inference costs.
DeepSeek-V3.2-Exp is "an experimental version of our model" that introduces "DeepSeek Sparse Attention — a sparse attention mechanism designed to explore and validate optimizations for training and inference efficiency in long-context scenarios," the company said in a post on Hugging Face. "Preliminary testing by DeepSeek found that the price of a simple API call could be reduced by as much as half in long-context situations," according to TechCrunch.
Bursting at the SIEMs: Microsoft's security software business is a rather large part of its enterprise tech portfolio, and this week the company added several new features to Sentinel, which falls into the SIEM (security information and event management) category. The Microsoft Sentinel data lake — introduced earlier this year — is now generally available, and it also added two buzzy AI-adjacent technologies to the platform.
Sentinel now supports (in public preview) graphing technology, and it will also soon add support for Anthropic's MCP protocol, which many vendors believe will help fulfill the agentic AI promise. Graph databases from companies like Neo4j played a key role in Klarna's decision to reduce its dependencies on external SaaS apps, and Microsoft thinks Sentinel graph will allow companies to "quickly reveal relationships, traversable digital paths to understand blast radius, privilege escalation, and anomalies across large, cloud-scale data sets," it said in a blog post.
Follow the data: As Microsoft's Sentinel team knows, the worlds of data analysis and cybersecurity are converging at a rapid pace. Databricks also knows a thing or two about data management, and this week it announced plans to get into the cybersecurity arena with Data Intelligence for Cybersecurity.
The new service "seamlessly integrates [YMMV] with enterprises’ existing security stacks, unifying all data and leveraging an open partner ecosystem so security teams can fully harness the power of AI — spotting risks earlier, understanding the full context of an attack and responding with greater speed," Databricks said in a press release. As Amazon's Steven Schmidt told Runtime in June, right now AI has been more of a help than a hindrance to cybersecurity teams, but that might not last forever.
Please check out the latest edition of the Runtime Roundtable, which this time around asked our distinguished panel of enterprise tech experts a very topical question: What is the best way to think about application deployment across cloud providers and self-managed data centers for both traditional and AI applications?
Thanks as always to our panel, and if you are interested in sponsoring a future edition of the Roundtable, please contact us here.
Everyone knows that enterprise AI is an expensive undertaking, but turns out most companies have no idea what they're getting themselves into. According to new research from Benchmarkit cited by CIO, "about 85% of organizations misestimate AI costs by more than 10%, and nearly a quarter are off by 50% or more," and believe it or not, they're not spending less than they anticipated.
“It really is that end-to-end experience for having an enterprise vibe coding for the agentic enterprise.” — Salesforce's Dan Fernandez, describing the company's Agentforce Vibes launch to TechCrunch in prose designed to sell bottles of whiskey.
Meta's zillion-dollar group of AI mercenaries is unhappy with the company's slow and awkward internal software development tools, according to Business Insider.
The hacking group behind the social-engineering attacks on Salesforce databases published a website Friday demanding payment to prevent it from leaking sensitive corporate data, according to TechCrunch.
Thanks for reading — see you Tuesday!
Today: The latest in the long-running saga of enterprise tech marketing departments trying and failing to look cool, Oracle customers are receiving extortion attempts after a breach, and the latest enterprise moves.
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.
Today on Product Saturday: Cloudflare gets into the data-management game, Glean unveils a new version of its AI work assistant, and the quote of the week.
Today: Edera's CTO explains how a rethink of hypervisor concepts could help companies protect themselves against software supply-chain attacks, Microsoft throws the Israeli intelligence service off its cloud, and the latest enterprise moves.
