Salesforce is building its AI product roadmap by polling its 18,000 customers in near real time, with some accounts meeting the company's engineering teams as often as once a week. The setup replaces the standard quarterly product cycle with a feedback loop measured in days, and Salesforce credits it for the pace at which Agentforce, its agent management platform, has shipped updates since launching in late 2024. The bet is that letting customers drive the backlog produces a roadmap that bends with the underlying AI tech rather than breaking against it.
Jayesh Govindarajan, executive vice president at Salesforce AI, told TechCrunch the 18,000-customer base is the input layer. "The 18,000 customers are a wellspring of information and a wealth of information that is really needed to get to customer success," he said. The company groups feedback by themes — agent context, observability, deterministic controls — instead of locking in fixed product timelines months in advance.
Muralidhar Krishnaprasad, president and CTO of Salesforce engineering, framed the shift as a forced response to how fast model capabilities are moving. "We can't wait three months or six months to get feedback, and then go figure out another six months of work," he said. "We are literally reacting to it, week by week, month by month." Code now ships behind feature gates so a subset of customers can stress-test new behaviors before a broad release.
Key facts
- 01Salesforce is crowdsourcing its AI product roadmap from its 18,000 customers, meeting with some as often as once a week.
- 02The company launched its agent management platform Agentforce in late 2024, ahead of the broader agentic AI wave.
- 03Engine's operations team holds weekly meetings with Salesforce and gets pre-release access to AI tools.
- 04PenFed built an ITSM workflow on Agentforce that Salesforce then rolled out to its broader customer base.
- 05Salesforce restructured teams to form a dedicated AI group when ChatGPT was released roughly a year and a half ago.
Salesforce was one of the first enterprise software vendors to put an agent management product into market, launching Agentforce in late 2024 before agents became the dominant industry storyline in 2025. Govindarajan said the platform exists because LLMs alone weren't enough — enterprises needed the last-mile plumbing to actually run agentic systems against their data and workflows. Salesforce has since extended into voice AI and Slack-native agents at a steady cadence.
“Salesforce is meeting with some of its 18,000 customers as often as once a week, replacing the standard quarterly product cycle with a feedback loop measured in days.”— Jaeden Schafer
Engine, a corporate travel management platform, sits inside that feedback loop. Founder and CEO Elia Wallen said his operations team meets Salesforce weekly and gets early access to tools before general availability. He recounted asking a Salesforce voice agent to book a Chicago hotel, finding the interaction stilted, flagging it, and then watching A/B test results improve after the agent was retuned.
"If somebody is willing to actually help curate and build products that we need, they can help us better and really understand our problem and how they can solve it," Wallen said. "For us, it's fantastic to actually be invited into a thing like that, because we can influence the product." The exchange runs both ways: Engine gets a competitive head start on tooling, and Salesforce gets a real production environment to validate ideas in.
PenFed, the federal credit union, is using the same arrangement to consolidate its software stack. Chief innovation officer and EVP Shree Reddy said his team built an IT service management workflow on top of existing Agentforce agents, and Salesforce liked the result enough to package the workflow for its broader customer base. "We invest our time, energy into the platforms that are more strategic," Reddy said, describing the relationship as mutually reinforcing.
Internally, Salesforce runs the same play. Govindarajan said employees are the largest users of the company's own AI tools, and Krishnaprasad said the company restructured headcount to spin up a dedicated AI team when ChatGPT first launched roughly a year and a half ago. "Agents weren't even in terminology when you look back a year and a half ago," he said. "We had to go react to all the advances, and we had to react to our customers."
The risk in this model is that the customer is not always right. Many enterprises are still figuring out where AI fits in their business, and a meaningful share have yet to extract durable value from agent deployments. Designing a product roadmap around what early adopters want this quarter can produce features that don't survive contact with a broader buyer base, or with the next model generation that obsoletes them.
There is also a commercial gap between piloting beta software in a weekly working session and signing a multi-year contract for it. Customers willing to co-develop are by definition the most engaged ones, which makes them a flattering but unrepresentative sample. If the median Salesforce customer never moves past experimentation, the roadmap risks optimizing for a vocal minority.
Still, the structural argument is hard to dismiss. Enterprise software vendors have historically planned in 12 to 18-month cycles, and that cadence is incompatible with a market where the underlying models change every few months. Salesforce is essentially admitting that nobody — including Salesforce — knows what the right AI product looks like in two quarters, so the only defensible strategy is to compress the distance between customer problem and shipped code.
For the broader enterprise AI market, Salesforce's approach reads as a tell about where leverage now sits. The companies most likely to win the agent layer are the ones with deep, daily access to how customers actually try to deploy this stuff in production — not the ones with the cleverest demo. That favors incumbents with 18,000 enterprise relationships over startups with better models but thinner distribution, and it raises the bar for any vendor still planning AI features on an annual roadmap.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
