Salesforce still dreaming about AI agents

Welcome to Runtime! Today: How Salesforce hopes to start generating real revenue from AI agents, why the merger of Fivetran and dbt Labs signals the end of an era in data infrastructure, and the latest funding rounds in enterprise tech.

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

Theater in the round

Few enterprise software executives have been more boisterous or evangelical about the power of AI agents to transform businesses than Salesforce CEO Marc Benioff, largely because there aren't many enterprise software executives who are more boisterous or evangelical than Marc Benioff. But for many reasons, agentic AI is still very much a work in progress.

Salesforce and Benioff sought to jump-start the agentic AI era Tuesday at Dreamforce in downtown San Francisco, one year after the entire enterprise software industry overwhelmed customers with a flurry of AI agent launches. The new plan is to put Slack front-and-center as the user interface to its agentic tools and allow its Agentforce platform to be a little more nimble when responding to customer prompts.

Salesforce customers will be able to access several key parts of the Agentforce platform directly through Slack's conversational interface, which the company called the "agentic OS for your enterprise" in a blog post.
That Agentforce platform is now known as Agentforce 360, and it comes with new tools for building custom agents as well as tweaking the responses produced by existing agents with Agent Script.
"Agent Script allows you to define agent behavior with a human-readable expression language, which enables conditional logic, precise tool use, and guided, deterministic controls," the company said.
That feature could give enterprises more confidence when deploying AI agents at key stages of their sales and customer-service pipelines.

During the keynote, as Benioff paced around the perimeter of a central stage name-dropping customers and partners, he claimed that Salesforce's own use of AI agents had saved the company $100 million and allowed it to respond to a much larger volume of inbound sales calls than it could ever handle with people alone. Those are the two central promises of agentic AI: we'll save you money and we'll make you money.

"That's why humans and agents have to drive customer success together so that we can get the customer loyalty we want, to get the trust that we want, to get the ability to have a leaner organization possible with this technology," he said.
A parade of Salesforce executives highlighted five customers who have adopted its agentic technology and, as you might expect from any vendor's biggest marketing event of the year, they had nothing but awesome things to say about it.
But PepsiCo's Athina Kanioura said that it doesn't expect to have completed its Salesforce-centric agentic AI push until the end of next year, which investors worried about Salesforce's revenue growth will surely note.
And in the closing minutes of the keynote, Benioff acknowledged that the company's pricing strategy for AI agents was not flexible enough for customers across its user base.

Perhaps inadvertently, toward the end of the morning Dell CEO Michael Dell laid out one of the main reasons why enterprise tech vendors have fallen all over themselves to embrace agentic AI. It's clear the technology has a lot of promise but is still too raw for most enterprises to embrace at scale, which isn't stopping vendors from urging companies to move more quickly by passing their own fears of disruption on to the market.

"We thought about [generative AI] as, 'well, if we don't really get on this, a new company is going to come along and it's going to put us out of business.'" Dell said.
Fear has been used to sell enterprise technology widgets since the dawn of electricity, but this time around the incredible hype surrounding generative AI technology stems as much from the existential fears of enterprise stalwarts like Salesforce and Dell as anything else.
That should give buyers something to think about as they're told to get on board with agentic AI or get left behind; who has more to lose, you or the person with an increasingly aggressive sales quota?

Extract and transform

It looks like the consolidation of the so-called "modern data stack" is underway, after Fivetran and dbt Labs agreed on Monday to merge in an all-stock deal, according to Reuters. The combined company would have around $600 million in annual revenue and will be led by Fivetran CEO George Fraser.

Fivetran and dbt Labs focused on slightly different parts of the data management pipeline, but in coming together they'll be able to defend against new incursions on their business models from the likes of Snowflake and Databricks. "The thing is really unique about this combination is our emphasis on open infrastructure and interoperability ... as everyone is trying to figure out how to use their business data in the context of AI," Fraser told Reuters.

It's yet another example of the "best of breed" strategy that allowed companies to mix and match different software tools across their businesses coming to an end amid increasing complexity and tighter budgets. According to Theory Ventures' Tomasz Tunguz, "as the cloud exploded onto the scene, the legacy data warehouse was replaced by a collection of fast-moving platforms. In that era, specialization won. The pendulum is now swinging back towards consolidation."

Enterprise funding

Reflection raised $2 billion in Series B funding as it attempts to build an open-source frontier model provider to challenge China's DeepSeek.

Reducto landed $75 million in Series B funding for its computer-vision technology, which turns unstructured text data into data businesses can use with large-language models.

Prezent scored $30 million in new funding and acquired another company co-founded by one of its co-founders.

Relace raised $23 million in Series A funding for its small AI models, which it believes will produce better coding agents.

Datacurve landed $15 million in Series A funding as it develops a data-quality platform designed to compete with ScaleAI.

Arcjet scored $8.5 million in Series A funding for its "local AI security model, an opt-in AI security layer that runs expert security analysis for every request entirely in your environment."

The Runtime roundup

Oracle patched another flaw in its E-Business Suite that was being actively exploited over the weekend with little notice, according to Bleeping Computer, which noted that it "reached out to Oracle more than six times for comment about the updates and the lack of disclosure regarding active exploitation, but received either no reply or they declined to comment."

Intel and AMD unveiled several enhancements to the x86 instruction set at the heart of their chips through a partnership between the two longtime rivals first introduced last year to ward off the competitive threat from Arm-based chips.

New data-center construction is becoming a flashpoint in Virginia politics after electricity bills increased over the last year, according to Semafor.

Thanks for reading — see you Thursday!