Salesforce Taps Customer Input to Shape Its AI Strategy

Artificial intelligence is advancing at a breathtaking pace, compelling businesses to develop and launch new products faster than ever to avoid being left behind by more agile competitors.

Salesforce believes it has discovered a strategy to stay ahead, even amidst the uncertainty of AI's future direction. The customer relationship management giant is crowdsourcing its AI roadmap in real-time.

While many companies seek customer feedback, Salesforce's approach stands out due to its scale, the rapid cadence of new releases and updates, and the depth of its partnerships. These aren't annual or quarterly check-ins; some customers engage with Salesforce as frequently as once a week.

“Our 18,000 customers are a vital source of insight, providing the rich information necessary for true customer success,” Jayesh Govindarajan, Salesforce AI's executive vice president, explained in a recent TechCrunch interview. “The platform we've built resonates with them. As we gather more context, our systems improve. With advancements in LLMs, agent systems are achieving increasingly autonomous behaviors. This is a long-term innovation journey, and we are committed to it.”

Salesforce was an early mover, launching AI agent management software in late 2024, before agentic AI became a major trend the following year. The company has since accelerated, consistently rolling out new products for voice AI and Slack.

Salesforce attributes its rapid release pace directly to its customers. By letting customer needs guide development, the company claims it can build an AI product roadmap that swiftly adapts to technological shifts.

When large language models first emerged, enterprises were eager to adopt the technology but lacked the crucial "last-mile" tools to implement them effectively, noted Muralidhar Krishnaprasad, President and CTO of Salesforce Engineering.

This gap in last-mile technology inspired Salesforce to launch its agent management platform, Agentforce, according to Govindarajan.

From there, the company adopted a bottom-up strategy focused on core themes—such as agent context, observability, and deterministic controls—rather than rigid product timelines. This method uses direct, ongoing feedback from select customer groups to build solutions, with the assumption that other enterprises will share similar challenges.

Customers Steering the Direction

“The innovations we've introduced are a direct result of collaborating with a wide range of customers and categorizing the real-world problems they face,” Govindarajan said. “We then analyze which issues can be solved at the LLM layer and which cannot. For those that can't, we build the necessary agentic operating system components around the LLMs to address them.”

Working closely with customers' engineering teams enables Salesforce to resolve issues quickly, before technology evolves and renders them obsolete.

“We can't afford to wait three to six months for feedback and then spend another six months working on a solution,” Krishnaprasad stated. “We are responding in real-time, week by week, month by month. This represents a significant shift. We now push code rapidly, utilizing various testing gates to trial new features and gather early feedback before a broad release. These are the adjustments we've made to thrive in this fast-changing environment.”

Engine, a travel management platform, is part of Salesforce's customer feedback loop. This is a serious commitment; according to founder and CEO Elia Wallen, Engine's operations team meets with Salesforce weekly.

Through this partnership, Engine gains early access to AI tools before public release. Wallen says this access helps Engine maintain a competitive edge and extract greater value from the technology.

The relationship is mutually beneficial.

Wallen has seen Engine's feedback directly incorporated into Salesforce's tools. For instance, after instructing an AI voice agent to book a Chicago hotel, he found the interaction somewhat unnatural and shared this observation with Salesforce. The agent was promptly refined, and subsequent A/B tests showed improved results.

“When a partner is willing to co-create products based on our needs, they gain a deeper understanding of our challenges and how to solve them,” Wallen said. “For us, being invited into this process is fantastic because we can actively influence the product.”

This strategy also allows Salesforce to identify and scale successful user-built solutions and workflows for its broader customer base.

Federal credit union PenFed has streamlined its technology stack through its close work with Salesforce, according to Shree Reddy, the company's Chief Innovation Officer and EVP.

“We invest our time and energy into strategic platforms, and we dedicate significant effort to this relationship with Salesforce,” Reddy commented. “This investment has strengthened our partnership, fostering mutual influence and delivering the greatest value for both organizations.”

Reddy explained that PenFed developed a custom IT service management (ITSM) workflow using existing Agentforce tools, which proved highly effective. Salesforce recognized this success and integrated the tool into its broader platform for other enterprises to use.

A potential downside to this approach is its reliance on the adage that "the customer is always right." Salesforce is betting they are, even though many businesses are still defining AI's role and have yet to see tangible value from the technology, which may not make them ideal guides for long-term product strategy.

Furthermore, a willingness to test beta technology now does not guarantee long-term adoption or future software contracts.

Being Your Own Primary User

Salesforce applies this bottom-up philosophy internally as well. Govindarajan notes that Salesforce employees are the most extensive users of its own AI tools.

The company also reallocated talent and resources at the dawn of the AI boom. Following ChatGPT's release, Salesforce restructured teams to form a dedicated AI unit—a strategy that has proven successful during previous waves of innovation, Krishnaprasad said.

“Technology is constantly changing; we can't predict what will emerge next month,” Krishnaprasad reflected. “We adapt. That's what we did throughout last year. Consider that the term 'agents' wasn't even common parlance a year and a half ago. We had to respond—to the technological advances and, most importantly, to our customers.”