This is why generic frameworks struggle here. We built our managed service on Provar to use metadata awareness instead of brittle selectors.

We handle the platform chaos, you get the green light.

springburst.io/salesforce-q...

Add in shared sandboxes: A single Validation Rule change by a dev can silently tank 50 regression tests.

This creates the illusion of randomness. You end up compensating with more hard waits and retries, creating a slow, brittle suite.

Salesforce is hostile to standard drivers:

IDs change at runtime (input-245 → input-246), breaking selectors.

UI renders after the DOM says "ready" (hello, async errors).

Shadow DOM acts like "privacy glass" for Selenium.

I wrote about what this shift means in practice: how we can test for reliability in autonomous systems.

🌿 Read: springburst.io/blog/testing...

#Salesforce #Agentforce360 #AITesting #QualityAssurance #AgenticEnterprise #Dreamforce2025

Agents now collaborate, make decisions, and sometimes… surprise us.

Testing is no longer about verifying workflows. It’s about validating trust, intent, and behavior.

That’s the new frontier for QA. 🌿

All of this inspired @dimitricalitri.bsky.social to this blog springburst.io/blog/salesfo...

A big thank you to Robin Gupta for the invitation, to TestZeus for sponsoring, and to Andrew Fawcett for sharing his valuable insights. A great start for the Humans of Testing series.

In our talk, Dimitri Fioole shared 3 practices that help keep orgs safe while projects stay on track:
🌱 Baseline regression packs
🌤 Preview sandbox validation
📅 Unified release calendar

Salesforce ships 3 releases a year. That cadence drives innovation, but it also puts pressure on testing. Both production orgs and projects in flight are impacted.

Full article here:
is.gd/genaitest

Testing generative AI is not about right or wrong.
It’s about understanding, relevance, and trust.

Start thinking like the history teacher you once had.

#GenerativeAI #SoftwareTesting #AIQuality #PromptEngineering #agentforce

So what do you do instead?
• Define what a “good” answer looks like
• Use semantic comparison, not string match
• Automate structure, not content
• Keep humans in the loop
• Track variability, not just accuracy
• Watch for tone, bias, and drift

Testing GenAI is more like grading essays:
• There’s no single right answer
• Meaning matters more than phrasing
• Human judgment is required

You need to test like a teacher, not like a script.

Ask your model the same question twice.
Get two different answers.
Both are valid.

That breaks traditional test automation, which expects 100% identical output.
But generative AI is built for variation, not replication.