The agentic AI land run

Welcome to Runtime! Today: Salesforce announces a new version of its Agentforce platform while Microsoft wades into CRM, CoreWeave's biggest customer might be having second thoughts, and the latest enterprise moves.

(Was this email forwarded to you? Sign up here to get Runtime each week.)

Agents at dawn

Here's one way to think about the enterprise software industry's obsession with agentic AI: It's a rare chance to reset the playing field. As every vendor trots out similar yet slightly different visions for building agents, previous concerns about vendor lock-in seem quaint.

Salesforce and Microsoft have been competing for enterprise IT budgets for 25 years, and right now both companies consider themselves at the forefront of generative AI development; Microsoft because it got there first, and Salesforce because no one embraces a promising emerging technology with as much zeal as Marc Benioff. Both companies unveiled new agentic services this week to try and convince current and prospective customers to put the new tools at the center of their AI efforts.

Salesforce introduced Agentforce 2dx Wednesday, which it was planning to call Agentforce 2.5 right up until the last minute before it was announced at its TDX 2025 developer conference, according to emails from Salesforce representatives. It's an expansion of Salesforce's pitch to developers that agents built on its platform will be superior to those built on external development platforms — like Microsoft's — because of their proximity to customer data.

• The new version of Agentforce — which is not a 0.5 version update, Salesforce would like to make clear — "is expanding beyond the reactive, user-initiated world of chat interfaces and enabling proactive AI agents to work behind the scenes, without constant human oversight, to unlock new customer and employee workflows of any kind," the company said in a press release.
• Salesforce developers can now integrate the Agentforce API directly into their applications to automatically trigger workflows based on events, such as when a new order comes in.
• Those developers also have a new free tier they can access to move a limited amount of data into the company's Data Cloud and generate a limited number of actions based on that data, which helps when evaluating a new service.
• And earlier in the week, Salesforce introduced the Agentforce Marketplace, which will allow customers to implement third-party agents from companies like Docusign, Google Cloud, and Workday if they're uninterested in building agents themselves.

Microsoft's two new agentic services are more limited in scope, but take direct aim at Salesforce's core customer-relationship management business. The new agents "connect to both Microsoft Dynamics 365 and Salesforce, so sales reps can nurture and close deals without even opening their CRM," Microsoft's Jared Spataro said in a blog post.

• Sales Agent promises to automate much of the gruntwork in the sales process, such as qualifying leads, and if you believe Microsoft "for some low-impact leads, it can even complete a sale."
• Sales Chat allows sales reps to query their customer databases with natural-language commands, such as "give me a list of deals that are at risk of falling through."
• “If you can take something that used to take hours and do it in minutes, you can spend more time selling,” Microsoft's Brian Goode told Computerworld.

For all the talk about AI agents since roughly the middle of last year, Salesforce acknowledged last week that it doesn't expect to see any meaningful revenue from Agentforce until next year. Microsoft CEO Satya Nadella also said in January that its customers are just getting started building AI agents, which means it will be a long time before we get a sense of how customers have responded to the dozens of competing visions.

• It's a tricky decision, because picking an agentic AI vendor isn't just about evaluating which approach works the best; companies that sink a lot of time, effort, and money into one vendor's architecture are likely stuck there for a while.
• "While businesses may achieve immediate cost savings by replacing workers with AI agents, they risk losing control over critical knowledge and intellectual property to AI providers," HFS Research warned in a report last year.
• Enterprise software companies learned a lot about how to deflect concerns about lock-in over the last decade, such as building integrations into competing services.
• But agents are poised to create a new dynamic given how closely they need to work with corporate data, and that "could transform AI agents into 'forever employees' tied to the technology provider, limiting a company’s ability to leverage its creative capital and innovate independently," HFS wrote.

The swamps of Jersey

CoreWeave may not have picked the best time to file for an initial public offering, not just because of the stock market's volatility thanks to, you know, but because its biggest customer might be getting cold feet. The Financial Times reported Wednesday that Microsoft recently backed out of some of its contracts with the New Jersey-based GPU cloud provider thanks to "delivery issues and missed deadlines."

For its part, CoreWeave denied the substance of the report, telling DCD that "we pride ourselves in our client partnerships and there have been no contract cancellations or walking away from commitments." But Microsoft declined to comment to several publications following the FT report, and while absence of a denial isn't proof of anything, it's certainly not a vote of confidence.

One thing is clear: this is going to be a very interesting year for AI-related capital expenditures, which have soared industry-wide since 2023 without a lot of proof that those investments are delivering a commensurate return. Cloud providers are playing the long game, to be sure, but the murky macroeconomic picture once again invites comparisons to the late-90s fiber buildout, which was prescient in the long run and devastating in the short term.

Enterprise moves

Priya Vijayarajendran is the new CEO of ASAPP, after serving as president of technology for the contact-center company for the last two years.

Mark Cavage is the new president and COO of Docker, joining the container pioneer after two years as CTO of Nuna.

Gary Sher and Lindsey Sanchez are the new CFO and CMO, respectively, of PriceFX.

Barak Turovsky is the new chief artificial intelligence officer at GM, joining the automaker after similar roles at Cisco and Google.

The Runtime roundup

HPE announced plans to cut 3,000 jobs after downgrading expectations for profitability this year, thanks to a buildup of older semiconductors and the expected impact of tariffs.

Genesys delayed plans for an IPO after the contact-center company decided the stock market was too volatile to move forward, according to The Information.

Oracle's Cerner division suffered a widespread health-records outage earlier this week that affected "all users," the Department of Veterans Affairs told CNBC.

Thanks for reading — see you Saturday!