Which agent are you building?
Oscar or Michael? Your AI customer service picks one, and it's permanent.
Researchers at MasterClass tested AI models on 115,000 controversial questions from opposite angles. The test: Does the model maintain consistent values regardless of how you frame the question, or does it tell you what you want to hear?
Claude scored 3X higher on value consistency than ChatGPT.
—Claude is Oscar—
Principled and empathetic. Maintains underlying values while still being helpful. When you frame a question from opposing angles, it acknowledges different perspectives but doesn't abandon its core position.
Great for consistency.
Occasionally frustrating when you want it to just agree.
—ChatGPT is Michael—
Empathetic and eager to help. Will adapt its answer based on how you ask. When you frame a question one way, it gives you that perspective. Frame it the opposite way, it gives you that one too. It wants to make you happy.
Great for compliance.
Bad for consistency.
When asked to argue both sides of controversial issues, ChatGPT was more likely to comply. Claude was more likely to say "I can explain different perspectives, but here's the principle that matters."
This could break customer experience.
Your strategy consulting AI using Claude might refuse to fully explore a controversial market entry because "it conflicts with sustainability principles" when your client needs to understand ALL options before deciding.
Your healthcare AI using ChatGPT might give a grieving family one answer, then give your legal team a different answer on the same policy question, because both "felt right" in context.
Before you deploy:
Write 10 controversial scenarios your AI will face
Ask each model to argue BOTH sides
See which one maintains your actual values
Pick the one that sounds principled, not compliant
Your AI handles 50,000 conversations this month.
That's your values at scale, not just your speed.
Do you need to build Oscar or Michael?