Salesforce Is Fixing One of AI’s Biggest Problems: Jagged Intelligence

AI is powerful—but too often, it’s unpredictable.

That’s the problem Salesforce AI Research is taking head-on: what it calls “jagged intelligence.” It’s the gap between what AI can do in theory—and what it actually does in messy, high-stakes business settings.

From hallucinated outputs to inconsistent performance, AI’s reliability issues have long been a roadblock for enterprise adoption. Now, Salesforce is building the roadmap to close that gap.

What Is “Jagged Intelligence”?

The term refers to AI’s uneven capabilities—it can ace one task, then completely flub a related one. This makes it risky for critical workflows where trust, accuracy, and repeatability matter most.

Think of an AI that can summarize legal documents flawlessly one moment—then misinterpret a contract clause the next. That’s jagged intelligence in action.

Salesforce’s Fix: New Models, Benchmarks, and Agent Frameworks

Salesforce’s approach is threefold:

• Custom enterprise models: Tailored to handle domain-specific tasks with higher fidelity and fewer hallucinations.
• Real-world benchmarks: Ditching synthetic tests for messier, business-aligned benchmarks that reflect how AI will actually be used.
• Agent frameworks: Built to reason across unpredictable tasks with memory, structure, and goal awareness—making them more like a trusted colleague than a fancy autocomplete.

As Silvio Savarese, EVP & Chief Scientist of Salesforce AI, puts it: “Our mission is to push AI from research labs into everyday enterprise operations—safely, reliably, and intelligently.”

Why It Matters

Enterprises want AI—but not AI that needs babysitting.

• Consistency > creativity: For sales, service, and operations, hallucinations aren’t just bugs—they’re business risks.
• Trust is everything: From financial audits to healthcare compliance, companies need AI that aligns with governance and legal standards.
• Agents are the future: The next wave of AI won’t be static models—it will be agents that interact, reason, and adapt across tools and teams.

Salesforce is moving to make those agents not just possible—but production-ready.

Key Players Driving the Push

• Silvio Savarese – spearheading the research and technical vision.
• Pascale Fung – focusing on trust and interpretability in generative models.
• Marc Benioff – doubling down on making Salesforce the AI CRM.

Their unified message? AI that dazzles in demos isn’t enough—it must deliver in daily enterprise life.

A Step Toward Trusted, Useful AI

With this move, Salesforce is aiming to lead the next phase of enterprise AI: where models don’t just wow—they work.

The industry is shifting from AI hype to AI hygiene: how to make it reliable, repeatable, and responsible. That’s where real value—and real adoption—will happen.

From jagged intelligence to trusted performance—Salesforce is drawing the line.

Sources: Salesforce AI Research blog