Runtime: How DataHub adds context; OpenAI expands Daybreak; AI heats up

Welcome to Runtime! Today: How DataHub helps companies building AI applications connect those apps with the right data, OpenAI loosens access (a bit) to its AI models for security tasks, and the debut of Stacked Up, a look at the week ahead.

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.

First up: Context clues

"Context" is quickly becoming the word of the summer in enterprise AI as companies building AI agents start to realize that they need data infrastructure to make sure those agents have everything they need to operate. Both Databricks and Snowflake devoted significant portions of their recent customer events to tools and services that can provide agents with that context, and the rest of enterprise tech is following suit.

But one of the most popular tools for finding context from within vast pools of corporate data is DataHub, which is available as an open-source project that is used by several household names in tech as well as a managed service from DataHub Cloud. Born at LinkedIn — just another needle-moving enterprise open-source project to emerge from that company alongside Kafka and Samza — DataHub "is one of the most mature attempts to take the metadata of an entire, fragmented data estate, apply familiar software design lifecycle concepts and governance, and create a metadata and context backbone for enterprises," Mary Branscombe wrote for The Stack.

As LinkedIn prepared for the rollout of the GDPR rules, it realized that it needed to find and connect data stored in several different places to comply with those regulations. DataHub consists of four layers of software that connect data with business context so that it can be quickly found when it is needed.

"Open source DataHub has what you need to run a real governance initiative: helping teams find data, catching downstream breakage before you ship it, and building the context layer agents are going to need," said Stanislav Dmitriev of Aiven told The Stack. It also helps reduce computing expenses given that agents will try and find a given piece of data no matter what the cost; telling it exactly where to go and what to get significantly improves velocity.

The rest of The Stack

Here come the daybreak: OpenAI divided its security-oriented Daybreak special-access program into two categories Monday, allowing Daybreak Blue participants to use its flagship models without the security guardrails that reduced their usefulness by a significant amount. Daybreak Red members will have far more unrestricted privileges when using those models as well as access to a brand-new security model.

The price is right: As more enterprises adopt AI only to recoil at the costs of using models in the apps, the frontier model companies seem to be getting the message. Anthropic will keep the pricing for its latest Sonnet model at its launch price despite earlier plans to raise it in September, weeks after OpenAI cut prices on some of its models.

Small but mighty: The UK Government Commercial Agency released the 15th version of the G-Cloud framework that public-sector agencies use to find and hire enterprise tech vendors without having to wade through a ton of bureaucratic steps. It's specifically decided to showcase small and medium enterprises, who traditionally "have faced barriers that hold them back from winning business - complex application processes, short framework windows, and procurement routes that favour incumbents," chief tech procurement officer Philip Orumwense said in a statement.

Join peers following The Stack on LinkedIn

Getting hot in here: The GPUs needed for AI workloads run much hotter than their CPU counterparts, and that has required data-center operators to redesign their buildings and equipment in order to deal with all that heat. As a result, business is booming for Irish building materials company Kingspan, which operates a division that makes floor and ceilings for data centers that can accommodate that heat and the cooling systems needed to make it go away.

Citation needed: Early experiments with using AI for legal purposes proved to be quite a disaster, but systems that can save lawyers from themselves are also emerging. In this week's edition of STACKUP, Noah Bovenizer talked to Hanna Roos, co-founder and CEO of Aavalynx, which is working on bringing AI tools into the dispute resolution process.

This could have been a fake meeting: Synthetic data is proving to be a surprisingly resilient way to train AI models, even in the public sector. The UK's AI and Data Readiness team recently discussed how they are using synthetic data to create fake emails, which were used to train an AI model that can sort through an ever-growing backlog of government email to understand which messages need to be kept and which can be discarded.

Stacked Up: Coming later this week

—CoreWeave reports earnings on Tuesday. Analysts expect revenue of ~$2.5 billion, which would be a 110% jump compared to last year, but they also expect its losses to widen.

—The AI Risk Summit takes place Tuesday and Wednesday in Half Moon Bay, Calif.

—Cerebras and Cisco report earnings on Wednesday. It's only the second-ever public earnings report released by the AI chipmaker, while Cisco gears up to report earnings for the 147th time (according to Google Search AI, anyway).

—Render(ATL) takes place Wednesday and Thursday in, of course, Atlanta.

—The 35th USENIX Security Symposium takes place Wednesday through Friday in Baltimore, Md.

The Stack Summit: We're convening in London on November 4-5, for a series of exclusive workshops and fireside conversations on the rise of BYOC as a favoured enterprise SaaS deployment model; how many CDOs are consolidating their data estates with Apache Iceberg; how CISOs at FTSE 100 scale are handling supply chain risk (with GSK's CISO) and more.

Ticket applications are subject to pre-vetting. Get in touch with ed@thestack.technology if you want to be in the room.

We're also reading:

“Code was never the hard part” is an insult to all programmers: A righteous rant from Senko Rašić pointing out that while AI coding tools are powerful, craftsmanship is not dead yet.

Thanks for reading — see you Thursday!