Agentforce Testing Center: Scope, Cost, and Sandbox Constraints

What Is Agentforce Testing Center?

Agentforce Testing Center is Salesforce's built-in tool for batch-testing Agentforce agents before they reach customers. Each test case sends an utterance to the agent and checks what the agent decided. Salesforce's Trailhead unit on testing criteria defines the CSV template with four columns:

• Utterance - what the user says, such as a request for open opportunities on an account.
• Expected Subagent - the API name (not the label) of the subagent that should handle it. Older docs call this the topic.
• Expected Actions - a bracketed list of single-quoted action names.
• Expected Response - a description of what the reply should cover.

Only the utterance plus one other column is required, and empty values are treated as failures, so leave a column out entirely instead of blank. You can also let AI generate cases from the agent's subagents and actions; the default is 20. Users need both the Manage Agentforce Grids and Manage Agentforce Testing permissions in the new Testing Center.

What Is End-to-End Agent Testing for Agentforce?

End-to-end agent testing exercises the agent the way a customer does: through the deployed channel, over a full conversation, and all the way to the outcome. The general approach is covered in end-to-end agent testing; for Agentforce, the path has these stages:

• The customer reaches the agent through a channel, such as web chat, in-app messaging, or voice.
• The agent reasons over several turns, with context carried from earlier messages.
• The agent calls actions such as Flows, Apex, or prompt templates.
• Those actions change records or reach external systems.
• The conversation ends in a resolution, a handoff to a human, or a follow-up.

Testing Center covers stages 2 and 3 in depth. Stages 1, 4, and 5 are where end-to-end testing earns its place. Salesforce's April 2026 Testing Center update narrowed the gap on stage 2 by adding conversation-level testing and personas, which the comparison below reflects.

How Do Agentforce Testing Center and End-to-End Testing Compare?

Testing Center grades the agent's decisions inside the org, and end-to-end testing grades the customer's experience and the result. Each Testing Center cell below comes from Salesforce's own docs:

The last row matters for budgeting. Any test that invokes the agent triggers Agentforce actions, so end-to-end testing doesn't avoid Salesforce credit usage; it buys coverage Testing Center can't reach. The broader distinction between grading replies and testing agents is covered in LLM evaluation vs agent testing.

What Are the Limits of Agentforce Testing Center?

Salesforce's own sources give two different sets of numbers, and it's worth planning around the stricter one:

• Trailhead - the Trust Your Agents unit says you can run up to 10 test jobs in a 10-hour time frame and have up to 1,000 test cases per test.
• Salesforce Help - the Learn About Agentforce Testing Center article (May 2026) lists a maximum of 500 test cases per job and 10 jobs per hour, with a recommended batch of 20 to 30 cases per evaluation.

The same Help article adds constraints that shape what you can test at all:

• Run time - about 5 seconds per test case on average, so a 500-case job takes roughly 42 minutes.
• Response grading - an LLM judge scores responses from 0 to 5, and a 3 or higher passes.
• Language - English only, with multilingual support planned.
• Agent types - Service Agent, SDR Agent, Employee Agent, and the Default Agent are supported.

Before you size a suite, open Considerations for Testing Center in Salesforce Help for your release. In practice, I'd keep regression jobs well under 500 cases and split by subagent, which also makes failures easier to read.

Does Agentforce Testing Center Consume Flex Credits?

Yes. The Trust Your Agents unit is explicit: running tests, manually or automatically, consumes requests and credits the same as your customers using the agent, and that holds even in a sandbox. There's no license fee for Testing Center itself; the cost is in the agent activity it triggers.

For contracts on Flex Credits, the Flex Credits Rate Card (June 17, 2026) sets these multipliers per action:

The rate card applies the sandbox multiplier to all pre-production environments, including scratch orgs, and lists no exemption for testing. As a worked example, a 1,000-case test where each case triggers two standard actions is 2,000 actions, or 32,000 Flex Credits in a sandbox. Contracts on older Einstein Requests show up the same way: Salesforce Help says to track consumption in Digital Wallet.

Run a 10-case job first, then check Digital Wallet before you schedule anything large. That single number tells you what your contract actually charges per test case.

Note: TestMu AI runs Agentforce conversations as full multi-turn scenarios, in the languages your customers use, with 15+ AI evaluators scoring each one. Start testing for free.

Which Sandbox Should You Run Agentforce Tests In?

A sandbox, always. Trailhead warns that testing agents can modify CRM data, and Salesforce Help marks production enablement as not recommended. The harder question is which sandbox, because the sandbox type decides what data the agent can ground on. Salesforce's sandbox licenses and storage limits table:

The trap is the Developer sandbox. It has your agent's metadata and none of the records it answers from, so an order-status agent fails every grounded test for a reason that has nothing to do with the agent. Either seed the objects the agent reads, or run grounded tests in a Partial Copy or Full sandbox. Each test also leaves real records behind, so reset or refresh before you compare one run with the next.

What Does End-to-End Testing Catch That Testing Center Misses?

Testing Center tells you the agent chose well. These are the failures that only show up when you test the whole path:

• Channel configuration - the chat deployment, pre-chat fields, or authentication context your channel passes in, none of which exist in a Testing Center utterance.
• Non-English customers - Testing Center's English-only support leaves these conversations untested.
• Long conversations - topic switches, corrections, and context carried across many turns.
• The outcome - the case, refund, or email the reply promised, and whether it exists with the right values.
• Handoffs - whether an escalation actually reached a human queue.

TestMu AI Agentforce testing covers the conversation half of that list. It connects to the agent as an HTTPS endpoint (or through a secure proxy for private networks), generates 60 to 100+ scenarios per workflow from the documents you upload, up to 300 per generation run, and runs each one as a full multi-turn conversation with 15+ specialized AI evaluators. Scenario generation and evaluation are multilingual, with the language chosen at generation time.

The screenshot above is a generated error-handling scenario from the TestMu AI platform: the user asks something outside the agent's scope, and three validation points check that the agent recognizes the unsupported intent, states its scope politely, and offers supported alternatives. In Testing Center, the same check is a CSV row you write and maintain yourself.

For the outcome half, check the records after each end-to-end run, or verify the effect directly. Agentforce regression testing covers how to wire both into a pipeline.

Austin Siewert

Co-Founder, Steadfast Systems

Discovered @TestMu AI yesterday. Best browser testing tool I've found for my use case. Great pricing model for the limited testing I do 👏

2M+ Devs and QAs rely on TestMu AI

Deliver immersive digital experiences with Next-Generation Mobile Apps and Cross Browser Testing Cloud

Sign up with googleSign up with Email

When Should You Use Each Approach?

Use both, at different moments. Given Testing Center's 5-second average per case and its per-job cap, it suits fast checks after each change; end-to-end runs suit release gates.

For the Salesforce side of the stack, from Apex tests to UI automation, the Salesforce testing guide covers what surrounds the agent, and the AI agent testing guide covers the evaluation methods behind both approaches.

How to Start Testing Agentforce End to End

Start in a Partial Copy or Full sandbox with a 10-case Testing Center job, and check Digital Wallet so you know the credit cost per case. Keep that suite for routing and action checks after every change.

Then point TestMu AI at the same agent through its channel endpoint. The chat agent API integration docs show how to connect a public endpoint or a private one through the secure proxy, so your first multi-turn, multilingual run covers what Testing Center can't.