“Vibe Coding” Is Real — Watch It Happen Live

Salesforce's Agentforce Vibes lets business users build AI apps through conversation, extending low-code development with natural language prompts grounded in company data

Forget syntax and specs. When Patrick Stokes unveiled Agentforce Vibes recently, the Salesforce EVP of Product & Industries Marketing had one word for the new technology: fun.

It's the fastest, and frankly most fun way to build anything just from a description.

"It's the fastest, and frankly most fun way to build anything just from a description," he said during the demo.

How does it work? Simply put, Agentforce Vibes reads context. This capability allows it to build applications from descriptions alone - no code required. "Context is what gives language meaning and understanding," Stokes explained. This ability for technology to understand the context of what a human is saying, makes it simple to turn ideas into apps, agents, and workflows.

This is made possible through "vibe coding" - a term coined by Andrej Karpathy in February 2025. The approach treats AI as a collaborative partner, allowing it to generate a first functional draft of code or components based on natural language instructions. Developers then fine-tune the output, shifting the focus from writing every line to shaping the overall solution.

Agentforce Vibes, which is now generally available, automatically connects to enterprise data in Salesforce. This helps the AI coding agent understand the language of your business and access the necessary context to support the output for any given prompt, such as building quality and scalable apps. Users have the ability to scaffold net-new custom objects, relationships, page layouts, and code Apex logic quickly.

And since vibe coding is as easy as writing something down, developers of any skill level can build functional enterprise code - fast.

"The AI is elevating the developer. And the AI is elevating every aspect of the enterprise," said Salesforce Co-Chair and CEO Marc Benioff.

Go deeper: