How Salesforce Built, Tested, and Scaled Our IT Support Agents

Across every industry and department, companies are onboarding an autonomous new workforce. AI agents are stepping in to handle complexity, speed up workflows, and redefine the modern workforce.

At Salesforce, we equip organizations with the tools to set their AI agents for success — grounded in your unique business context and designed with built-in guardrails. 

But we aren’t just building these too. We consider ourselves, Customer Zero. This means we are our own first customer, prototyping the path so you don’t have to. We hold ourselves to the highest standards to ensure that by the time a solution reaches you, it has already delivered real impact for us. For us, Agentforce is a pivotal solution for how every company can become an Agentic Enterprise.

Here’s the story behind building IT support agents to ensure that technical glitches don’t stand in the way of a productive workday.

The legacy process

Behind every great company is an IT team working tirelessly to keep things running. At Salesforce, that team is called “Techforce.” But even for a global tech leader, managing 25,000 support tickets a month for 76,000 employees is no small feat. From lost laptops to access resets, the volume was high.

Customer success starts with our employees. For example, if a developer is stuck on an IT ticket or an HR question, they aren’t building the software that helps our customers succeed. Along with our team of human agents, Techforce used a chatbot for basic troubleshooting. With the volume of monthly tickets, however, we needed a more sophisticated solution — and Agentforce was the answer.

Once implemented, the results were instant:

• 40% of IT support cases are now handled by agents.
• $57,000 in cost savings was achieved within the first two months.
• 9,500 cases have been resolved automatically.

How did we get from experimentation to deployment? Let’s retrace our steps.

Behind the scenes of building Techforce agents

To get our AI agents ready to serve our employees and support IT teams, we use the same developer toolbelt we provide to our customers: Salesforce Sandboxes, Data Mask & Seed, and Backup & Recover.

Building and testing high–performing agents

Every high-performing agent, app, and workflow begins its journey in a Salesforce Sandbox. Salesforce offers four types of sandboxes: Developer, Developer Pro, Partial Copy, and Full Copy. Each has unique storage and refresh limits, so choosing the right sandbox environment is key to an efficient dev cycle.

Salesforce developer teams deployed three Full Copy Sandboxes. Two replicated Agentforce Service (formerly Service Cloud) — one for User Acceptance Training, and the other for auto deploy and debugging — and one replicated Data 360. Because Full Copy Sandboxes are exact replicas of the production environment, it is optimal for rigorous end-to-end agentic functionality testing

Fully Copy Sandboxes are especially essential for agent development because: 

• RAG and data grounding: Agents rely on Retrieval Augmented Generation (RAG). To test if an agent is pulling the correct context, it needs access to the actual volume and variety of your production data. Without a Full Copy, you risk “hallucinations” in production that never appeared in testing.
• Vector database accuracy: Testing how Data 360 indexes information requires a production-scale dataset to validate that search queries and semantic clusters are performing accurately.
• End-to-end agentic logic: Agentforce doesn’t just follow a script, it reasons. Testing that reasoning requires the same complex web of historical cases, customer profiles, and legacy records the agent will encounter in the wild.

Ensuring secure, seamless user experiences

Trust is our #1 value at Salesforce, so while our agents were getting smarter, we needed to ensure our data stayed secure.

To protect our team and meet global standards, we’ve operate with the highest data security and privacy guardrails. Using Data Mask & Seed, we obfuscated PII and sensitive project records in all three Full Copy Sandboxes before they ever touched the “home org.” This verified that while our agents were learning our business context, sensitive details like home addresses and private IT logs remained strictly protected within the companion org throughout the entire testing lifecycle.

Once rolled out, we focused on creating a secure, frictionless user experience by enforcing the principle of least privilege. By migrating to a new Agentforce Service environment specifically to eliminate “birthright access” for non-essential users, it reserved direct permissions only for those handling active cases. Extending this functionality through Slack and Experience Cloud enabled internal teams to interact with Agentforce safely across the entire business — from sales and marketing, to the front lines of Techforce.

At Salesforce, we believe in building custom employee apps that are tailored to the specific role. Instead of providing broad, over-provisioned access to the entire CRM, you can build streamlined experiences that match the actual job to be done. This ensures every employee has exactly the tools they need to be productive—and nothing they don’t—all while maintaining a clean and secure environment.

Protecting the blueprint 

While Full Copy Sandboxes provide the perfect playground for Agentforce, they also represent a significant investment in configuration and data preparation. To protect this “pre-production” reality, a comprehensive backup and recovery strategy was essential. 

With Salesforce Backup & Recover, the team was able to implement:

• An “Undo” button for AI experiments: Agentic workflows can be unpredictable during the tuning phase. If a prompt or logic change causes an agent to erroneously update thousands of records during a test, a quick restore allows the team to “rewind” the sandbox without a manual cleanup.
• Metadata integrity: Agentforce relies on complex metadata—permissions, flow logic, and prompt templates. Backing up this metadata ensures that if a deployment goes sideways, the “gold standard” configuration can be recovered instantly.
• Seeding and refresh cycles: Full Copy Sandboxes are refreshed periodically. A dedicated backup solution allows the team to use a backup snapshot to capture specific “test states” or datasets and move them between sandboxes, ensuring consistency even when the sandbox itself is reset to match production.

The Salesforce stamp of approval

Agents aren’t siloed — they are integrated across our entire business. Our sales agents automate lead-to-cash, our marketing agents drive pipeline, and our Techforce agents have streamlined everything from engineering tasks to IT support.

As Customer Zero, we are building the future of Salesforce from the inside out. We’re happy to do the heavy lifting of testing and launching products internally first, so you can trust they are ready to scale when they reach you.

Ready to build your own Agentforce? Learn more about AI agent development on Salesforce.

Attending TDX 2026? Get a behind the scenes look at Salesforce’s agent and application development cycle in the How Salesforce Builds Agents Securely in Sandboxes session.