AI Testing · 7 min read

How We Get to ~90% AI-Generated Test Cases

Inside our prompt library and the Provar MCP / Playwright MCP workflow — what it drafts, what humans still own, and why we stop at 90%, not 100%.

Every Salesforce QA team we've worked with has the same hidden cost: writing the test case takes longer than running it. A well-structured Provar or Playwright test case — steps, test data, assertions, and edge cases — routinely takes a skilled engineer one to three hours to author by hand. Multiply that by every flow, every CPQ rule, every release, and authoring becomes the real bottleneck, not execution.

What AI Actually Generates

Our prompt library takes three inputs: Salesforce org metadata (objects, flows, validation rules), user stories or requirements, and — where available — historical defect data. Those prompts run through Provar MCP or Playwright MCP, which draft a complete test case candidate:

  • Ordered test steps that match the actual flow logic in the org
  • Test data mapped to real field names and picklist values
  • Assertions tied to expected outcomes at each step
  • Edge cases surfaced from flow branches and past defect patterns, not just the happy path

In practice, this covers about 90% of what a finished test case needs. That number comes from measuring how much of an AI-drafted case survives engineer review unchanged versus how much gets corrected or added.

Why We Don't Automate the Last 10%

The remaining 10% is where judgment lives: ambiguous business rules that aren't fully captured in metadata, compliance-sensitive paths where a wrong assumption is expensive, and genuinely novel edge cases — the ones nobody has hit yet. AI can flag these ("this branch has no historical test coverage") but shouldn't resolve them unsupervised. That's the work our Provar and Playwright engineers do on every case before it ships.

What This Changes for Your Team

It doesn't remove your QA engineers — it moves them up the stack. Instead of typing test steps, they're reviewing AI drafts, deciding what the edge cases actually mean for the business, and owning the parts of the suite that require real domain judgment. The result: suites get built in days instead of weeks, and the humans on the project spend their time on the 10% that actually needs them.

See It on Your Org

We can run this against a real flow from your Salesforce org during the free 10-hour trial, so you can see exactly what gets generated and what our engineers change before it ships.

Watch AI Draft Your Next Test Suite

Send us a flow from your org — we'll show you the AI-generated draft and the engineer review, side by side.

Start Free AI Trial — 10 Hours →