I usually start by testing different inputs, including unusual ones, and checking whether the code handles errors properly. I once accepted an AI suggestion that worked perfectly with sample data but failed when some fields were missing. Since then, I also check variable names, dependencies, and whether the solution fits the rest of my project.