A passing test can repeat the same mistake as the code.

If an implementation and its test share the same wrong assumption, their agreement may leave the important question untouched.

AI assistance can make both easier to produce. The team's testing ability still depends on knowing what behavior matters and which conditions would challenge the apparent success.

That is a useful place to bring different expertise into the work.

Ask where the expected behavior came from.

Is it grounded in the user's purpose, an operating requirement, an established interface, or an interpretation nobody has examined? Which participant could identify a condition the current example leaves out?

These questions help separate testing the implementation from testing the underlying account of what it should do.

They are especially useful when a prototype has encouraged several groups to expect different results.

Use the question behind the test.

What observation would make the team less confident? Where does the system cross a boundary? Which behavior matters when the convenient conditions disappear? What information would help someone diagnose the result?

A test becomes more informative when the group can explain why the observation matters.

This can guide productive use of AI assistance: generate candidate cases or explore a behavior, then examine their relevance to the actual system.

Make expert knowledge part of the experiment.

An operator, experienced developer, domain specialist, or product colleague may recognize a condition the immediate task does not reveal.

Howard Pearlmutter's work includes distributed-system diagnosis, integration, technical qualification, prototypes, and practical AI-assisted development. His experience can help connect a test question with the mechanism and purpose it needs to examine.

Read the Telos engagement and Howard's background for relevant examples.

Start with a result the team trusts for different reasons.

Ask participants why they believe a piece of work is ready. Compare the observations behind their confidence. The differences can reveal a useful next test or a missing shared definition.

A workshop can make that discussion concrete. Mentoring can help individuals carry the reasoning into their own work.

The aim is greater testing ability: more informative questions, better use of experience, and a clearer understanding of what success actually establishes.