How Salesforce Uses 18,000 Customers to Build Its AI Roadmap in Real Time

With AI evolving faster than any traditional product cycle can absorb, Salesforce has taken an unconventional approach to staying competitive: it’s handing its AI roadmap directly to its customers. The company runs a continuous, real-time feedback loop with 18,000 enterprise clients — some meeting with Salesforce teams as often as once a week — and uses those conversations to drive product development across its Agentforce platform.

Salesforce was one of the first companies to launch AI agent management software in late 2024, before agentic AI dominated headlines. The company has since doubled down, releasing new products for voice AI and Slack at a pace its CTO attributes directly to customer feedback.

Customers in the Driver’s Seat

Rather than building to specific product timelines, Salesforce organizes development around operational themes — agent context, observability, and deterministic controls — derived from direct customer experience. EVP of AI Jayesh Govindarajan told TechCrunch: “The innovation that we’ve brought, they are the direct result of us working with a vast number of these customers and then classifying the problems they see in the real world.”

The relationship is genuinely bidirectional. Engine, a travel management platform, meets weekly with Salesforce and reported that feedback about an AI voice agent feeling “unnatural” was implemented quickly — with A/B tests showing measurably better results shortly after. Federal credit union PenFed built a custom IT service management workflow inside Agentforce that Salesforce subsequently rolled out to its broader enterprise customer base.

Key Takeaways

• Salesforce is reacting “week by week, month by month” — CTO Muralidhar Krishnaprasad noted the company must react not just to customer needs but to AI advances that didn’t exist “a year and a half ago”
• The company shifted labor and resources when ChatGPT launched, creating a new AI team — a strategy it has used successfully during past technology shifts
• The risk: enterprises still figuring out what role AI will play in their business may not be the best source for long-term product decisions, and beta adoption doesn’t guarantee future software contracts

🔗 Read the full article on TechCrunch