Databricks has bad news for SaaS

Welcome to Runtime! Today: Databricks co-founder Reynold Xin on the speed at which AI-driven coding is changing software, Snowflake launches a coding agent and previews its Postgres database, and the latest funding rounds in enterprise tech.

Please forward this email to a friend or colleague! If it was forwarded to you, sign up here to get Runtime each week, and if you value independent enterprise tech journalism, click the button below and become a Runtime supporter today.

Become a Runtime Supporter

The soft thing about soft things

As software stocks continued to get pummeled this week by some combination of investors convinced the AI-coding revolution is nigh or desperate to lock in their gains, count Databricks' Reynold Xin in the former camp. While the data analytics company's customers tend to be on the bleeding edge of technology trends, they're sending a strong signal that a new era of software development is well underway.

"We actually think the world will shift from a proliferation of SaaS software to proliferation of bespoke, custom software, largely coded by AI and operated by AI," Xin said in an interview with Runtime this week. Databricks' co-founder and chief architect is far from the first person to point out that trend, but that shift is happening so quickly that Databricks changed some of its development priorities behind its Lakebase database, which was announced last June at the company's Data & AI Summit and became generally available Tuesday.

Lakebase is the product of Databricks' acquisition of Neon, a managed PostgreSQL database company, in May 2025.
Last year CEO Ali Ghodsi described Lakebase as a new take on the venerable transactional database, one that was designed to scale up and down quickly with demand by separating compute and storage in a serverless architecture.
The idea at the time was to provide a better developer experience for companies working with older transactional databases and create a foundational layer for agentic AI, but the company soon realized the ground was shifting underneath those developers much faster than anyone realized at last year's event.

In the second half of 2025, Databricks customers started building way more custom internal applications than they had ever considered even scoping out thanks to the rapid improvement of AI coding tools, Xin said. All those applications need databases, yet Neon wasn't designed to operate across an enterprise that might need thousands or even "millions" of databases to serve those apps in this new world, he said.

"Historically, databases were designed to be managed one at a time, but no company runs one database," Xin said; most companies "run a small fleet of databases" and employ several people to manage those databases..
That strategy simply won't work in a world of mass-production software, but companies still need "a single pane of glass" — a dashboard, really — to manage the reliability and security of all those databases, he said.
Databricks realized that "once you scale up your database management, you should start treating database management as an analytics problem," Xin said, which as it turns out is a problem that Databricks' flagship product is pretty good at solving.
The company shifted development priorities to address that management layer and will add additional features in subsequent releases.

Xin cautioned that right now this explosion in internal enterprise software development is unevenly distributed, pointing out that it took nearly a decade before cloud computing became mainstream. But "the cost of software development is going down dramatically and will continue to shift down," he said, which changes the math on how companies will allocate resources.

The tricky part will be figuring out how to run and maintain all those applications, which is just one reason why companies — especially companies that aren't in the software business — paid SaaS vendors a lot of money over the last decade to take that burden off their plates.
Still, a shift toward building away from buying "is an enormous opportunity for infrastructure, because that would mean there will be a proliferation of infrastructure to support all these applications, whether infrastructure databases, hosting of those applications, or model inference," Xin said.
It will also mean that infrastructure tools will need to be designed for people without years of experience in database or infrastructure management, because there simply aren't enough people to manage that many apps using older tools designed for subject-matter experts, he said.
That's why SaaS vendors are working very hard to adapt to this new era, because you're not paranoid if they really are out to get you.

Slippery slopes

Snowflake is also seeing a lot of the same trends as its fierce rival Databricks, which likely timed the GA release of Lakebase to tweak Snowflake at its Build 2026 conference in London. Snowflake announced several new additions to its data platform Tuesday, including a new AI coding tool as well as its own foray into transaction processing.

Cortex Code was designed to help users build data pipelines or run data analytics projects using natural language commands, and it works either directly within Snowflake or within development tools like Visual Studio Code or Cursor. The company also announced that Snowflake Postgres — its answer to Databricks Neon/Lakebase project — will be "generally available soon" for "applications that require high-throughput transactions and large-scale analytics simultaneously."

"With the PostgreSQL service, our goal is to provide this secure boundary where, if customers build apps or build agents within that boundary, their data has not left the compliance and regulatory perimeter for Snowflake," executive vice president Christian Kleinerman told The Register. If vibe coding apps are freaking out enterprise software investors, Snowflake and Databricks' determination to own a wider variety of database workflows should freak out dozens of special-purpose database vendors.

Enterprise funding

Decagon raised $250 million in Series B funding, which values the agentic customer-support company at $4.5 billion.

Gruve landed $50 million in a follow-on to an earlier Series A round to expand its AI inference services, which use software to run inference jobs across a network of data centers.

Adaptive6 launched with $44 million in seed and Series A funding as it builds out a cloud cost-optimization platform that scans cloud infrastructure deployments for spending waste.

Outtake raised $40 million in Series B funding for its digital trust technology, which helps companies protect themselves against identity-based attacks and credential theft.

Orion scored $32 million in Series A funding as it builds out a data-loss prevention platform designed around custom large-language models and AI agents.

Day AI landed $20 million in Series A funding to build a new type of CRM application designed around AI agents.

The Runtime roundup

Oracle announced plans to take out as much as $50 billion in debt to finance its AI data center buildout, and investors were not impressed.

Wall Street also didn't like AMD's earnings beat and increased guidance for the current quarter, but it was a brutal day for tech stocks overall.

Microsoft's Managed Identity for Azure service took much of the evening off on Monday, which affected several other Azure services that rely on that tool, according to The Register.

United Airlines plans to upgrade its reservation system during the wee hours of the East Coast morning Wednesday, so if you're flying United tomorrow, good luck.

Thanks for reading — see you Thursday!