We are fully in the agent era, and the question on every business leader’s mind is the same: where is the return on our AI investment?
94% of enterprises report they still are not realizing a strong return on their AI investments. The gap lives in what comes after the agents are built, when teams try to operationalize them. The challenge is scaling agents safely, connecting them to the systems that run your business, and keeping costs from spiraling. That is what we are addressing at Dreamforce this week.
Building agents was the easy part
When it comes to scaling agents, businesses have become exposed to three gaps that don’t show up in pilots:
- Lack of visibility and control: agents sprawled across platforms, built by different teams, with no single view of what they’re doing or who owns them.
- Limited agent action: agents have the ability to answer questions or make recommendations, but not to act in the systems where real work actually happens, leaving employees bogged down in the routine tasks AI was supposed to eliminate.
- Spiralling AI cost problem: token consumption compounding month over month with no clear line back to business value.
Getting a real return on AI means solving all three, and that’s what Agent Fabric helps you achieve.
Govern every agent no matter where it was built
Most organizations are discovering that the number of agents running across the enterprise is almost certainly larger than anyone expects. That agent sprawl creates a compliance and security liability. The organizations moving confidently into production are the ones that have established centralized governance, applying consistent policies across every agent regardless of where it was built. Governance is what gives teams the confidence to experiment and then scale fast.
Put agents to work in the systems that run your business
Governance creates trust. But a trusted agent that cannot reach the systems where your business actually runs will struggle to deliver value. Agent Fabric is built on the integration, API management and ‘connect anything’ leadership of MuleSoft which serves as the perfect foundation. Agent Fabric helps you achieve trusted action quickly by enabling you to extend your existing APIs and integrations into agent workflows, so agents can take action without rebuilding from scratch.
Connectivity gets agents to work. Keeping them working is a different discipline. An agent that hits the right outcome once and quietly breaks a week later is a liability, not an asset. That is the problem our newest announcement Agent Evals is built to solve. Before you scale an agent, you need to know it is consistently doing the right thing.
Agent Evals gives teams a systematic way to test agent behavior at scale, auto-generating test scenarios, scoring outcomes against defined success criteria, and surfacing failures before they reach production. Because evaluations live on the same platform as governance, a failing agent can be tied directly to a remediation action. That is what turns a promising pilot into something the business can depend on.
Make the economics of AI work at scale
Governed agents delivering consistent outcomes can still break the program if the economics do not hold. Token costs are adding up fast, and most organizations find themselves without clear visibility into what is driving the spend or how to control it.
Sustainable return on your AI requires real-time visibility into consumption, hard controls that stop overruns before they happen, and intelligent routing that matches model cost to actual task complexity. Every token should be traceable back to the value it creates.
Hands-on innovation across the campus
For those at Dreamforce this week, the MuleSoft Campground at Moscone South, the Data Lab for Trusted Insights, and the IT Lounge are running hands-on demos, expert sessions, and practitioner meetups across all three of these areas. Come see these ideas working in practice.
