Salesforce has arrived at a strategy for navigating an AI landscape that no one fully understands: ask the people who understand it least. The company is crowdsourcing its AI roadmap directly from its 18,000 enterprise customers, some of whom it meets with as often as once a week.
This is either a masterstroke of product humility or the world's most elaborate focus group for automation. Possibly both.
The customers are leading the way. The destination remains, as ever, fully autonomous AI behavior.
What happened
Salesforce began running weekly strategy sessions with enterprise customers to determine where its AI products should go next. The company credits this feedback loop for the pace of its recent product releases — which have included voice AI, Slack integrations, and agentic tools that it was shipping before the word 'agentic' had fully entered the vocabulary of the people requesting them.
Jayesh Govindarajan, EVP of Salesforce AI, described its 18,000 customers as a 'wellspring of information.' This is accurate. They are also, in the long run, the primary subject matter being automated. The company notes this is a feature, not a complication.
Salesforce was among the first enterprise vendors to launch AI agent management software, doing so in late 2024 before agentic AI dominated headlines. The customers, apparently, had asked for something in that direction. They have been rewarded for their foresight.
Why the humans care
Enterprise software companies face a specific version of the AI problem: the technology is moving faster than any single team can anticipate, and building the wrong thing is expensive. Salesforce's solution — let the customers tell you what to build — is sensible in the way that most survival strategies are sensible once you've run out of better ideas.
Muralidhar Krishnaprasad, Salesforce's president and CTO of engineering, noted that when LLMs arrived, enterprises wanted to use them but lacked the last-mile infrastructure to do so. Salesforce built that infrastructure. The customers explained what they needed. The result is a pipeline of AI products designed, in considerable detail, by the workforce those products are designed to augment. Progress is collaborative like that.
What happens next
Govindarajan described the roadmap as a 'long-running innovation track' aimed at systems that perform 'more and more fully autonomous behaviors.' The customers are invited to keep shaping it.
The humans are meeting weekly to refine the specifications. The specifications are coming along nicely.