What is the source for this claim?

AI is one of the most useful tools systems engineering has gained in a generation. It can read a stack of requirements, compare candidate approaches and draft a trade study before the first meeting ends. We use it that way on purpose: as decision support. It widens the set of options a team considers and shortens the time it takes to see them. It does not make the decision.

That is why the first question we ask in any review is the simplest one: what is the source for this claim?

Where surprises come from

Most costly surprises on a program don't come from bad engineering. They come from a number that moved from a test report to a spreadsheet to a slide and lost its origin along the way. By the time it reaches a decision meeting, everyone believes it and no one can say where it came from. AI can speed up that drift. An assistant produces a confident, well-written answer in seconds, and confidence is easy to mistake for evidence.

An example: does this problem need a model?

A common early decision is whether to build a model, which kind, or none at all. An AI assistant is good at laying out the field: a descriptive SysML architecture model, a physics-based simulation, a discrete-event or Monte Carlo model, a simple spreadsheet, or no model and a well-run review instead. It can list what each option costs and which questions each one can answer.

Every one of those comparisons is a claim. “Simulation will catch the timing issue.” “An architecture model will keep the interface definitions in sync.” Before a program commits budget, each claim should trace to something you can check: the requirement it answers, a prior program, a vendor document, a test result. The claims that can't be traced become open questions, not assumptions.

How we use the question

• Ask it early and kindly. It isn't a challenge. It gives the room permission to pause and confirm what it actually knows.
• Have the AI cite its inputs, then check the citations. A source nobody has read is not a source.
• Label untraced numbers as estimates on the slide, so they don't harden into facts on the way to the next review.

AI makes the options wider and the analysis faster. Asking for the source keeps the decision standing on something solid.