L03 Brief: The Big Three: Context, Model, Prompt
EECS AASE Podcast, season 1, episode 3

Transcript of the published audio. The episode is AI-generated from
course materials; this transcript is produced locally from the audio.

This is the brief on Lecture 3 of EECS 498 AASE. I know you're itching to
fire up your IDEs, but consider this your wake-up call. Today introduces
a career-long diagnostic framework where you'll learn that when the AI
spits out absolute garbage, it's almost certainly a user error, not a
model error.

First, we're diving into the big three levers for every single session.
Context for what you show it, model for which one you use, and prompt for
how you actually phrase your instructions. So why is this a bullseye and
not just some simple checklist? Well, if you miss even one lever, no
amount of effort on the other two is going to rescue your code.

Second, you'll see that out of those three, context is the one that
causes the absolute most trouble. It's the lever you control most
directly, and honestly, it's the main culprit behind those frustrating
"the AI got it wrong" moments. You'll be tempted by the trap of thinking
more is better, which is a debate you're definitely going to have with
your neighbor today, but dumping your entire code base in there is
actually worse than carefully selecting the right things.

Finally, you've got to stick around for the end of class where we turn
debugging into a live spectator sport. We're running a single task on a
five-file code base using Aider, watching it fail in real time with the
critical file missing, then running it again with the file added, proving
that context is a deliberate engineering decision, not just a passive
setup step.

Mastering AI coding isn't about crossing your fingers and hoping the
model works. It's about intentionally engineering your context, your
model, and your prompt.