AI agents are now capable of managing and completing complex workflows, with or without human … [+]
This morning Salesforce launched AgentExchange, a marketplace where developers can create and sell AI agents, and businesses can browse, test, and buy pre-built AI actions, templates, and agent solutions. Launching with “more than 200 initial partners” and “hundreds of ready-made solutions,” AgentExchange will help businesses build out an AI agent workforce to complement and enhance their living and breathing employees. And, maybe, replace them.
“This is going to be the last generation of managers who manage a wholly human workforce,” Salesforce AI executive Alice Steinglass told me last week while sharing the news. “We just launched AgentForce and are seeing tremendous growth and success with 3,000 deals already … customers want this new form of digital labor.”
AgentForce is Salesforce’s digital labor platform designed to help businesses build, deploy, and manage AI-powered agents that automate tasks and workflows. AgentExchange is a marketplace on top of AgentForce for browsing and buying ready-made usable AI agents.
Agents are AI-driven automation components that are built to handle business tasks like processing data, executing workflows, helping customers find products, or solving common tech support problems. They can help sales teams track leads, schedule follow-ups, generate proposals, or automate contract signing. But they can be adapted to almost any business need. One use case Salesforce told me about is what medical device maker Neuron7 is doing with agents: diagnosing the problem and figuring out the next best action when a medical device needs to be serviced.
Agents are one of the hottest spaces in AI right now, given that they don’t just follow standard repetitive automation tasks but can accomplish things that until now have required human intervention.
“Automation is great when the tasks are exactly the same,” says Steinglass, who is an EVP and general manager of platform at Salesforce. “Agents are for more complex tasks … a customer can define the outcomes and the goals and let an AI reason through that.”
Salesforce isn’t the only company building and selling a platform for agents that can accomplish business tasks. In 2024 competitor Hubspot built and launched agent.ai, a network for agents where businesses can “discover, connect with, and hire AI agents.”
In a year it has grown to 1.2 million (human) users, 9,000 of which have actually built an agent, Hubspot CTO Darmesh Shah told me last week. According to the agent.ai website, however, only 919 of those have been made publicly available for people to use and buy.
“This is the year of agents,” Shah says. “For the first time, the LLM can do orchestration and executive planning.”
From Salesforce’ perspective, agents are in some sense the new apps. For years the company has had AppExchange, where service providers and entrepreneurs could built apps for Salesforce customers, and those customers could simply add it to their Salesforce instance immediately: sort of an Amazon-style one-click shopping experience for 9.5 million B2B apps.
Salesforce sees the same thing happening with agents. But the difference is the kinds of jobs that AI agents can now accomplish: read web pages for information, create images to help deliver information, sense tone and emotion in a support request, or complete complex multi-step projects like doing background research on a person or a company.
Another key difference: who can create agents.
“Anyone can build an agent,” says Steinglass. “Salesforce has always been good at low-code/no-code, and low code enabled a new class of builders. AI will do even more.”
(Low-code and no-code are ways of creating applications without programming, or with minimal software development.)
Steinglass says anyone will be able to talk to AI, tell it what they want, and get their agent automatically built.
Which brings up the prospect, of course, of rogue agents. Imagine smart AI agents that do the wrong thing, either intentionally or unintentionally. Salesforce says they’ve built a safe sandbox for agents that will stop any nefarious activity.
“You need to have guardrails and protections in place,” Steinglass says. “That’s what we’ve built at every layer of the platform. Users have rules; agents need the same set of rules.”
Salesforce is also building tools to help manage digital team members. Managers need to see what their direct reports are working on, Steinglass told me, and it’ll be no different with digital workers. Agents are currently handling 40% of the questions on Salesforce’ help website right now, and there’s a seamless transition to human workers when needed.
Which brings up a question: will this cost human jobs?
Hubspot’s Shah says yes, at least initially:
“In the interim, we’re going to see some disruption in the labor force,” he told me. “You’re going to see some roles reduce. But we’re going to see more entrepreneurship … you’ll be able to do with a two-person company that might have taken 20 people years ago.”
The common saying these days is that AI won’t take your job, but a human who uses AI might. That’s probably small comfort to someone who does routine tech support for a living, and to support their families. But there is likely no way to stem the flood of AI that we’re seeing injected into all facets of business and personal life right now.
And the benefits are real and undeniable.
I asked Steinglass what the person she was five years ago would say when seeing all the capability that AI is now providing.
“I got into this job because I love technology,” she said. “Back when I was learning to do AI in the 90s, it couldn’t do much and it was still amazing. At every point, it surprises me what we can do … generative AI and what we can do right now is the biggest surprise of my career.”
