L05 Brief: Modern agentic CLIs
EECS AASE Podcast, season 1, episode 7

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 5 of EECS 498 AASE. Imagine for a second
that the familiar AI tool you already use is actually a fully agentic
system, because today we're exploring the massive shift that happens when
a model starts choosing its own next steps. First, let's break down the
four absolute primitives that make a tool truly agentic. A loop, tools,
planning, and memory. And here's the wild surprise. Aider, the AI coding
assistant you might already know, literally scores a perfect 4 out of 4.

But if Aider has every single piece of an autonomous agent, why doesn't it
actually feel like one? You know, it's kind of like sitting in a high-
performance sports car, but you're still stuck manually working the
clutch yourself. Second, we're bringing this theory straight back to your
own workflow during a mid-lecture chat break. Get ready to look at your
everyday setup in a totally new light, as we take the autonomy gradient
and apply it directly to the exact tool you unpacked in Lab 1. Finally,
we're gonna step

up that autonomy ladder with a live, completely unscripted demonstration
right on the podium screen. You'll watch the instructor run the Claude
Code model on the taskr platform, and no, you absolutely do not need any
accounts or subscriptions to follow along. You'll witness a really
crucial tension play out, because while the model picks its own actions,
the software harness still strictly dictates which actions are even
possible. Ultimately, moving up the autonomy ladder isn't just about
unlocking new capabilities, it's about the reality of handing over the
steering wheel.