We stress-test our software on the most demanding, high-maintenance customer we could find: ourselves. 75,000 Salesforce employees run our products in our real jobs before those products reach you. We file tickets. We query in multiple languages. We break things no roadmap anticipated. We call ourselves Customer Zero. Here are four things we've learned by being our own beta testers: 1. Start where nothing exists yet - Pick an operational area with a genuine business need and no current solution. No process to disrupt, no legacy baseline to defend. If the pilot stumbles, you haven't damaged trust in a system people already depend on. 2. Know your human baseline - We don't expand an agent past its release ring until it clears 70% accuracy, measured against the people it's replacing. Then we make it harder, writing new test cases from real support data. Last year's excellent becomes this year's floor. 3. Earn the easy wins first - Our internal IT agent started with the simplest job available: answering knowledge questions. It handled 30% of them almost immediately. That early result bought the credibility to attempt the complicated workflows. 4. Build into the flow of work - Put agents where people already spend their time. If adoption depends on changing daily habits, it won't happen. Which of these four is hardest where you work? The receipts: - Our internal IT agent climbed from a 76% satisfaction score to 90%. That didn't happen overnight. It happened over 18 months of subjecting it to our own real-world problems. - Our Help Agent has handled more than 4 million conversations and resolves close to 70% without a human. Now we're targeting 80% by the end of the year. - A device-matching problem our own employees exposed became Agentforce Graph, which customers now get out of the box. The full story on Customer Zero: