I’ve also been teaching my agentic setup to recognise when I’m drafting posts like this, which has helped streamline things.
One trick I rely on heavily is asking meta-questions first. Instead of jumping straight into a task, I’ll ask an agent how it understands a workflow or a piece of code. That way, when we reference a file or concept later, we’re already aligned.
Sometimes the most powerful prompt is just dragging a file into an AI and asking:
“What is this? How does it work? Do you see any errors? Is there a reason this might fail?”
I’ve been doing that almost daily for years. There are simply too many file formats and systems to learn exhaustively. This approach works better than Googling, where you’re hoping the top result matches your context. With AI, you get something closer to consensus reasoning instead of keyword matching.
People can argue about AI information quality all they like, but realistically most of us were already learning from Reddit threads, forum posts, and half-remembered tutorials. This just makes the process faster—and more honest.
Cheers, Adrian