Salesforce handed part of its AI roadmap to customers and sped up releases of Agentforce and new features

Salesforce is changing the logic of releasing AI products: instead of a long-term plan, the company is building a roadmap together with customers and rewriting priorities almost in real time. This approach should help it faster adapt Agentforce, voice AI, and Slack integrations to the real needs of large corporations.

Customers instead of a plan

The pace of AI development forces corporate vendors to release features significantly faster than before. At Salesforce, they decided that in such an environment, a classic quarterly roadmap works too slowly. That's why the company relies not so much on fixed deadlines, but on a constant stream of customer feedback. According to Salesforce AI leaders, some customer meetings happen every week, not once a quarter.

For Salesforce, this is not just conventional customer development. The company believes that if one large organization encounters a specific problem when implementing AI, similar pain will soon emerge at others. Internally, this becomes a system for early detection of needs: the team watches which problems repeat and turns them into product themes for the entire platform.

How the cycle works

The driver for launching Agentforce was simple: after large language models emerged, companies lacked the "last mile" to bring AI to a working enterprise scenario. The model alone is not enough—context, control, observability, and clear rules are needed so the agent doesn't behave unpredictably. That's why Salesforce builds the AI stack from the bottom up and plans not specific release dates, but directions of development.

• context for AI agents and access to company data
• observability—the ability to track what the agent does
• deterministic controls—restrictions and predictable rules
• early beta tests and "gates" before wide release
• fast code release cycle by weeks and months, not semesters

"We literally respond week after week, month after month," said

Muralidhar Krishnaprasad, president and CTO of Salesforce's engineering division, describing the new work rhythm.

The next step is to break down each customer request into parts: what the LLM layer itself can solve, and what requires additional agent infrastructure around the model. This is important for an enterprise environment where a beautiful demo feature is not enough. You need a product that can be controlled, verified, and integrated into an existing stack without constant manual oversight.

First results and risks

One of the participants in this cycle is the travel platform Engine. Its team meets with Salesforce weekly and gets access to AI tools before public launch. According to Engine's CEO, this helps the company test new capabilities earlier than competitors and really influence the product. One example is a voice agent for hotel booking: after noticing that the dialogue sounded unnatural, Salesforce refined the agent's behavior, and A/B tests showed better results.

A similar story happened with PenFed. The credit union built its own ITSM workflow based on already existing Agentforce tools and agents. When the solution proved itself in real work, Salesforce brought this scenario to the platform as a more universal tool for other customers. That is, customers here act not only as testers but also as co-authors of finished enterprise features.

This model has a downside. It assumes that the customer understands well what AI they will need a year from now, although many companies are still only looking for practical returns from such systems. Willingness to stay in beta and discuss the product every week also does not guarantee a long-term contract or mass adoption. Salesforce reduces this risk by actively using its AI tools internally and constantly redistributing teams to new technological shifts.

What this means

Salesforce shows what enterprise-AI development could look like in an era when the market changes faster than quarterly planning. The winners here will not be those with the most beautiful roadmap slide, but those who faster turn recurring customer pain points into working products for the entire customer base.