Salesforce: AI agents will show you the money

Welcome to Runtime! Today: Salesforce jumps on an AI bandwagon with its trademark enthusiasm, Apple reveals more details about its secure private cloud infrastructure, and the latest enterprise moves.

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

Stop saying "operator"

In the years since establishing itself as a cloud enterprise software pioneer, Salesforce has scrambled to position itself at the vanguard of just about every trend that has captivated Silicon Valley. Its thirsty pursuit of the generative AI boom is just the latest example, and next week at Dreamforce CEO Marc Benioff and the "ohana" plan to talk a lot about AI agents.

In a briefing for the media Thursday, Benioff and Salesforce AI CEO Clara Shih unveiled Agentforce, a collection of AI software that promises Salesforce customers an easy way to create chatbots and customer-service drones that can resolve common customer interactions far more quickly and cheaply than clunky phone trees or an army of actual people. In a rambling preamble to the Agentforce rollout, Benioff sought to position "agents" as a more refined version of the "copilots" that longtime rival Microsoft has made the center of its product strategy over the last two years.

• Unlike services such as ChatGPT that predict more-or-less correct responses to common queries, agents are empowered to take real action based on the inputs they receive.
• Salesforce created several agents that it says can handle typical business functions such as customer service, sales qualification, and marketing-campaign evaluation within flagship products such as Sales Cloud and Service Cloud.
• Agentforce will also help customers build their own custom agents based around their Salesforce data through a drag-and-drop user interface.
• "Our core vision is to help our customers connect with their customers in a whole new way," Benioff said.

The pitch for AI agents is that they can respond to a variety of inquiries with the speed and accuracy of a human being. "When it comes to agents, one of the most exciting capabilities is not just answering questions, but being able to automate business logic and actions," Shih said.

• Agentforce works by drawing upon the "Atlas Reasoning Engine," which Salesforce said "is built on a proprietary system designed to simulate how humans think and plan," without much elaboration on what that actually entails.
• But the concept is that Atlas can comb existing data on previous customer interactions and execute far more tasks than a "press 1 for men's clothing" style interaction could ever hope to handle.
• A live demo showed how those users can program their agents with links to internal data to respond to questions such as "is a different size of that sweater available in a local store" with a few clicks.
• The underlying message, of course, is that Agentforce users won't need as many human customer-service representatives to handle basic customer inquiries.

Agents are widely viewed as the next step in the generative AI revolution; former Salesforce co-CEO Bret Taylor's first act upon leaving the company was to found a startup called Sierra that does pretty much exactly what Agentforce promises to do. But the question remains whether any of this technology is actually good enough to put at the heart of business interactions.

• Very few people enjoy navigating automated phone trees, as this great Verizon commercial lampoons, so it's not hard to see how agents could improve those interactions.
• But as an event held this week by Theory Ventures' Tomasz Tunguz illustrated, nobody really knows how AI agents could impact the broader business world.
• "Only VCs care about the word agents / agentic : for most enterprises, agents mean a customer support agent," Tunguz wrote.
• If that remains true, the billions being currently invested in enterprise generative AI technology could have a pretty narrow impact.

Thinking different

During the iPhone years, Apple has mostly thought of the cloud as a storage closet, rather than an application platform. However, the rise of generative AI applications that require hefty processing power simply can't run on an iPhone, which led Apple to announce its Private Cloud Compute strategy earlier this year.

On the heels of Apple's annual September iPhone event, Wired got a detailed look behind the scenes at how Apple put together PCC's unique approach to cloud security. PCC runs on custom servers powered by Apple's own processors and a new operating system that "incorporates hardware and software security features the company has developed for Macs and iPhones over the past two decades," according to Wired.

The goal is to create a way for iPhones to take advantage of AI features running on remote servers without making the security compromises that become possible when personal data leaves a device. Other cloud and chip companies have built similar architectures, but Apple says it has gone a step beyond other approaches and will make "every production PCC server build publicly available for inspection so people unaffiliated with Apple can verify that PCC is doing (and not doing) what the company claims, and that everything is implemented correctly," Wired reported.

Enterprise moves

Carolina Dybeck Happe is the new executive vice president and chief operating officer at Microsoft, a newly created role for the former GE CEO that reports to CEO Satya Nadella.

Jill Popelka is the new CEO of Darktrace, after serving as the company's COO for the last five months.

David Roberts is the new CEO of SugarCRM, replacing Craig Charlton, who will become chairman of the sales and marketing software company.

Abhijit Mitra is the new CEO of Outreach, replacing new executive chairman Manny Medina after a year leading product and technology at the company.

Barbara Larson is the new CFO of SentinelOne, joining the security company after a decade at Workday.

Scott Meyer is the new CTO of ID.me, following more than a decade at LinkedIn.

The Runtime roundup

OpenAI released a new group of models called o1, which are said to "reason" better than its previous models but which the company said are still in preview mode.

Crowdstrike executives ignored quality control concerns in the years leading up to the devastating July outage, according to developers interviewed by Semafor.

Meanwhile, Microsoft will encourage security software companies to build products that work outside the Windows kernel, but stopped short of vowing to close off access to the kernel.

An AT&T glitch prevented users of its network from accessing Microsoft Azure and Microsoft 365 for several hours Thursday morning.

Thanks for reaching — see you Tuesday!