Lab activity · Agent design
Spot the critical error.
Ten short descriptions of AI agents at work. Each has one serious design flaw: the kind that turns a useful tool into an incident. Pick the flaw, then see why it matters.
The common thread
None of these ten flaws is about the model being clever enough. Each one is about authority, checks or evidence.
These ten flaws come from the failure modes in a working draft of an agentic-systems competency framework that Maverick is developing. It is not a certification, and this isn't an assessment. It's a quick way to see where agent designs usually go wrong.
Runs entirely in your browser and sends nothing anywhere. The only thing remembered, in this browser, is which activities you have finished.