Artificial intelligence is evolving at a breakneck pace, pushing companies to roll out new products faster than ever or risk falling behind more agile competitors. Salesforce believes it has found a way to stay ahead, even when the future direction of AI remains uncertain. The customer management software giant is effectively crowdsourcing its AI development roadmap in real time. While Salesforce is far from the only company that works closely with customers for product feedback, its approach stands out given the company’s massive scale, the rapid cadence of new product launches and updates, and the depth of its customer relationships. These aren’t occasional check-ins held annually or quarterly. Salesforce meets with some customers as often as once a week. “The stack we’ve built has resonated with these customers,” a company representative noted. “Over time, we can gain context to improve, and as it gets better—and as LLMs improve—agent systems will handle more fully autonomous behaviors. That’s a long-term innovation track, and we’re going to invest in it.”

EMBED_PLACEHOLDER_0

Salesforce was among the first companies to launch AI agent management software in late 2024, before agentic AI became a dominant topic the following year. Since then, it has doubled down, rapidly releasing new products for voice AI and Slack. The company credits its customers for driving this pace of innovation. The need for that last-mile technology sparked the creation of Agentforce, Salesforce’s agent management platform, according to Jayesh Govindarajan, executive vice president at Salesforce AI, in a recent interview. From there, the company adopted a bottom-up strategy organized around themes—such as agent context, observability, and deterministic controls—rather than fixed product timelines. This approach relies on direct feedback from rotating groups of customers to build products, with the assumption that other enterprises will face similar needs. Customers are in the driver’s seat.

“The innovations we’ve brought are a direct result of working with a large number of these customers and classifying the real-world problems they see,” Govindarajan said. “Then we break that down and determine what can be solved at the LLM layer and what cannot. For things we can’t solve at the LLM layer, we need to build agentic operating system components around the LLMs to address them.”

Working so closely with customers’ engineering teams allows Salesforce to fix issues quickly before technology evolves past them. “We can’t wait three months or six months for feedback, then spend another six months on work,” said Krishnaprasad. “We’re reacting week by week, month by month. That’s been a big change. Now we push code quickly, with various gates to test new features and get early feedback before a broad release. These are all changes we had to make to keep up with this fast-moving environment.”

EMBED_PLACEHOLDER_1

Engine, a travel management platform, is one of the companies in Salesforce’s customer feedback loop—and the relationship is far from casual. Engine’s operations team meets with Salesforce weekly, according to founder and CEO Elia Wallen. Through this partnership, Engine gains access to AI tools before they’re publicly released. Wallen says this early access helps Engine stay competitive and extract more value from the tools than it otherwise could. The relationship is reciprocal. Wallen has seen feedback from Engine implemented into Salesforce tools. For example, he once instructed an AI voice agent to book a hotel in Chicago. He felt the voice and interaction were a bit unnatural and shared that with Salesforce. Shortly after, the agent was updated, and the company’s A/B tests began showing better results. “If someone is willing to help curate and build products we need, they can understand our problem better and solve it more effectively,” Wallen said. “For us, it’s fantastic to be invited into something like that because we can influence the product.”

This strategy also allows Salesforce to roll out solutions and workflows built by users to its broader customer base. “We invest our time and energy into platforms that are more strategic, and we obviously spend a lot more time on this relationship,” said Reddy about Salesforce. “That investment has yielded good results in strengthening the partnership, influencing each other, and providing the best value add for both organizations.”

Reddy noted that PenFed developed an IT service management (ITSM) workflow on its own using existing tools and agents in Agentforce, which worked well for the company. Salesforce saw that success and rolled out the tool to the broader platform for other enterprises to use. The downside of this approach is that it relies on the classic service sentiment that the customer is always right. Salesforce is betting on that, even though many enterprises are still figuring out what role AI will play in their business and many have yet to find value from the technology. As a result, customers might not be the best source for long-term product development. Additionally, being willing to test and preview beta technology doesn’t necessarily translate into long-term usage habits or future software contracts.

Be your own biggest user

The company also applies this bottom-up approach internally. Govindarajan said Salesforce employees are the biggest users of its AI tools. At the start of the AI boom, the company shifted labor and resources. When ChatGPT was released, Salesforce moved teams and resources to create a new AI team—a strategy it has found successful during different innovation waves in the past, according to Krishnaprasad. “As technology changes, we never know what’s going to come out a month from now,” Krishnaprasad said. “We’ll adapt to it. That’s what we did all last year. If you think about it, agents weren’t even in the terminology a year and a half ago. Then we had to react to that. We had to react to all the advances and to our customers.”