L04 Brief: Spec-Driven Coding
EECS AASE Podcast, season 1, episode 6

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 4 of EECS 498 AASE. Get ready, because the
fundamental bottleneck in AI-assisted work has officially moved. You see,
writing the actual code just isn't the slow part anymore. Saying
precisely what you want is. First, we're tackling this planning
bottleneck head-on. In 2026, just throwing clever prompts at an AI? Yeah,
that doesn't really cut it. Instead, we're shifting to the discipline of
plan equals prompt.

Think of it like this. You're no longer the bricklayer hammering out
syntax. You are the architect, simply handing over the blueprints.
Second, once you've got that architect mindset, we look at the spec
document. This replaces your basic prompt with something way more
powerful. A spec uses information-dense keywords, always pairing an
action with a specific location, so the model literally never has to
guess.

We're talking clear file boundaries, strict start and end contexts, and
mapping out tasks in exact dependency order. And hey, you'll actually get
to test drive this. Right in the middle of class, we're taking a chat
break where you'll use this exact vocabulary to outline a feature
yourselves, though I'm definitely not spoiling what you're going to
build. Finally, we wrap up with a live demo of Aider running in architect
mode.

It does this in two steps. It drafts a plan in prose, and then it just
stops and waits. So why introduce a pause to wait for a human? Well,
because reading a prose plan is the absolute cheapest place in the entire
workflow to catch an AI that misunderstood you. Mastering this spec-
driven coding is the ultimate discipline you absolutely need to nail down
before diving into Lab 1 and your graded build.