The World Cup is on, and fans across the globe are rooting on their favorite players. They thrill every time Messi scores a goal, Díaz completes a daring pass, or Martínez blocks a shot. And yet ultimately, no matter how brilliant an individual performance, all that truly matters is whether the team won.
Agents today are brilliant individual performers. They resolve cases, qualify leads, answer calls, and with tools like observability, you can even understand how effectively they accomplish those objectives. You can even coach them toward better performance. Resolving a customer’s issue, a better-qualified pipeline: those are outcomes, and getting agents to drive toward them is real progress.
That’s loop engineering at work. Rather than walking an agent through a script of prompts, you hand it an objective and a way to measure progress, and it does more than finish a task: it plans, checks its own work, learns from the result, and adjusts. For clear, individual objectives this already works remarkably well, which is why a coding agent can land a record number of accepted pull requests.
But an individual outcome is still an individual stat. More pull requests don’t tell you whether the company shipped a better product and sold more of it. The ultimate goal isn’t completing a task. It’s moving the business forward, and that’s harder to measure.
This is the real frontier: the same loop, aimed higher. Self-learning and optimization across the entire experience, with sales, service, and marketing pulling toward the same business outcome instead of each agent optimizing its own corner. But that’s only possible if you can measure that outcome.
Here’s the good news for our customers: You have already solved this problem. For twenty-seven years you have encoded your goals into Salesforce. How you create and close revenue. How you earn return on marketing. How you take care of your customers. Pipeline stages, service levels, qualified lead definitions, all recorded in the platform that runs your business every day. And all that CRM data, customer signals, automation, and analytics turn out to be the perfect infrastructure to translate tasks into goals that agents can optimize against over time. There’s no need to build a new scoreboard, because the business itself is the scoreboard.
We also give the agent the four systems it needs to reach its goals: the controls to govern what it’s allowed to do, the context to ground it in your business, the traceability to see every step it took, and the analytics to know whether the outcome actually moved. Together, those systems provide the harness that turns a running agent into a learning one. You can see where it fell short, change the system that produced the miss, prove the change is better, and carry it forward. We call this the Agent Development Lifecycle (ADLC), and we’ve been building this way since before the industry had a name for it.
This isn’t theoretical — it’s the head start our customers already have. Take Zing Health, the Medicare Advantage insurer. Every month, Zing fields more than 150,000 member calls, benefits and drug coverage questions too intricate for traditional IVR. When they built Mia, their AI agent on Agentforce Voice, it didn’t start from a blank slate: the member records, plan configurations, and service histories it needed were already on the platform. Today Mia delivers instant, bilingual support across 33 plans, with intelligent routing and seamless human handoff, because the business logic that defines good service at Zing was encoded long before the agent arrived.
Telepass, Italy’s mobility leader, shows the same thing at scale. Processing 1.4 billion transactions a year, they deployed a customer-facing FAQ agent and an internal support agent in just weeks, drawing on the customer records and service configurations that already lived in Salesforce. These agents now handle 40,000 conversations a week, with 87% of them resolved without any human involvement.
Those are strong agent-level outcomes. But the real advantage shows up a level higher, when agents work together across the whole platform. Indeed shows what that looks like. The hiring marketplace runs four Agentforce agents — from an SDR agent that books meetings to internal agents that support employees. None of these agents runs in a silo: they sit on the same customer record, the same platform, so an outcome in one corner is visible to the rest. That is the end-to-end view almost no one else can offer: when your sales, your service, and the work of your own people all run on one system, an agent optimizing for the business can finally see the whole field.
First we built agents that could answer. Then agents that could act. Now we’re building agents that pursue a goal on their own: running the loop, reading their own results, and adjusting. You can see the shape of it in third-party projects like OpenClaw and Nous Research’s Hermes Agents that don’t wait to be retrained or reprompted, but read their own results and decide what to do next. The difference, when that frontier arrives in the enterprise, is that an agent learning on Salesforce isn’t just optimizing its own performance, but moving the entire business forward.
So take your cue from the World Cup teams we’re seeing this summer, teams that don’t just know their task, but can read the whole match and adjust in pursuit of the win. Your agents need that same clarity. You bring the goal, and Agentforce keeps the score. Let them play to win.