Salesforce is building its AI products by sitting in weekly meetings with customers and shipping fixes within days, according to TechCrunch AI. The company has flipped the traditional enterprise playbook, replacing multi-quarter roadmaps with a bottom-up loop where 18,000 customers feed problems directly into engineering. TechCrunch AI reports that this approach is now the engine behind Salesforce’s rapid push into agentic AI, voice agents, and Slack-based tools.
What stands out here is the cadence. These aren’t quarterly business reviews or annual feedback sessions. Salesforce engineers are meeting with select customers as often as once a week, watching them use pre-release tools, and pushing code in response.
The strategy: themes, not timelines
Jayesh Govindarajan, executive vice president at Salesforce AI, told TechCrunch that customers are “a wellspring of information” guiding what the company builds next. Instead of locking in product specs months ahead, Salesforce organizes work around themes like agent context, observability, and deterministic controls. The bet: if a problem shows up in one customer’s stack, it’ll show up in many.
Muralidhar Krishnaprasad, president and CTO of Salesforce engineering, framed the urgency directly. “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.”
This is significant because most enterprise software vendors still operate on legacy release cycles. Salesforce is essentially admitting that nobody knows where AI is headed, so the only viable strategy is to react fast and let users surface the gaps.
Customers as co-builders
Two customer stories in the TechCrunch piece show how this plays out:
- Engine, a travel management platform, meets with Salesforce weekly. CEO Elia Wallen told an AI voice agent to book a Chicago hotel, flagged that the interaction felt unnatural, and watched Salesforce ship a fix that improved A/B test results shortly after.
- PenFed, a federal credit union, built its own IT service management workflow on top of Agentforce. Salesforce liked it enough to roll the pattern out to its broader customer base. Shree Reddy, PenFed’s chief innovation officer, said the relationship has let the credit union slim down its tech stack and influence the platform directly.
Why this matters now
The AI infrastructure war isn’t just about model quality anymore. It’s about the “last-mile tech,” the orchestration, observability, and guardrails that make LLMs actually usable inside an enterprise. Salesforce launched Agentforce specifically because customers wanted to use LLMs but didn’t have the plumbing.
The risk is real, though. This model assumes the loudest, most engaged customers represent the broader market. If Salesforce optimizes for the 50 enterprises in the room, it could miss what the other 17,950 actually need. There’s also a classic agency problem: customers ask for what they want today, not what they’ll need in 18 months.
Practical takeaways for AI builders
If you’re building AI products in 2026, the Salesforce playbook offers a few signals worth copying:
- Shorten the feedback loop. Quarterly user research is dead in AI. Weekly is the new floor.
- Plan in themes, ship in slices. Locking specs months out is a losing bet when models change every quarter.
- Pick partners, not just users. A handful of deeply engaged customers will outperform a thousand survey respondents.
- Build observability early. You can’t react to feedback you can’t measure.
For enterprise buyers, the read is different. Vendors willing to bring you into the cockpit are worth more than vendors with prettier roadmaps. The question to ask isn’t “what features are coming?” It’s “how fast can you ship when I tell you something’s broken?”
Expect more of the big AI platforms to copy this model in 2026. The companies that figure out how to crowdsource a roadmap without becoming hostage to it will own the next phase. Full reporting available at the original TechCrunch AI source.