A1.01 · Why an Agile Agent Workflow
The two failure modes
Two habits sink most software built with an agent. Prompting without a plan yields code no one specified and no one can accept; writing one enormous specification yields a document no one finishes. This module names both failures precisely, then the unit that escapes them.
Three dives, one point each on the same curve: the no-plan failure (vibe coding), the over-plan failure (the big-bang spec), and the thin, vertical, provable slice — specified only enough to be built and accepted — that the rest of the course is built on.
Too little plan, too much plan
Plot a way of building a rung against how much is settled before any code is written, and the failures sit at the two ends. With too little — prompting an agent and keeping whatever compiles — the result is code no one specified and no one can sign off: vibe coding. With too much — one exhaustive specification written before a line of code — the document is never finished, drifts from a reality it never met, and returns no feedback until it is too late to use: the big-bang spec.
The work that ships lives between them. A thin slice carries only enough plan to be built and proven — one vertical cut of real value, accepted, before the next begins. It is the same shape every later rung in the course takes, and the unit the whole workflow exists to produce.
Risk of failure · against how much is planned up front
Both extremes fail: no plan ships the unspecifiable, all plan ships nothing. Value is delivered in thin, provable slices.
The first rung is not "the store" but one branded id, specified and proven, then shipped — the next slice built on an accepted one.
Neither no plan nor all plan ships reliable software; the unit that does is a thin slice of value, specified only enough, proven before the next begins.
The three dives
Each dive takes one point on that curve. The first two name a failure mode precisely and show it on the Portal; the third is the slice that resolves them — the shape every later rung takes.
Vibe coding — the no-plan failure
Prompting without a spec or acceptance: unreviewable diffs, no definition of done, and entropy that compounds rung over rung. Why the code that compiles is not the code you can keep.
Big-bang specs — the over-plan failure
The exhaustive specification written before any code: never finished, drifting from reality, and silent on whether it works until the end. Why more plan is not more certainty.
Thin slices — the resolution
One vertical cut of real value, INVEST-sized and tracer-bullet-shaped, specified only enough to be proven and accepted. The unit the whole workflow produces, built first on the Portal.
After A1.01, the next module — A1.02 · Pragmatic Programming, revisited for agents — re-reads the pragmatic canon for a world where an agent writes the code. Return to the A1 module navigation to continue.
References
The canon this module distils, and where its slice fits in the course.
Sources
- Hunt & Thomas — The Pragmatic Programmer — tracer bullets and walking skeletons: the thin-slice alternative.
- Beck, K. — Extreme Programming Explained — small batches and short feedback loops over big up-front design.
- Cohn, M. — User Stories Applied — INVEST and the thin, vertical slice of value.
Related in this course
- A1.01.1 · Vibe coding
- A1.01.2 · Big-bang specs
- A1.01.3 · Thin slices
- A0.2.1 · The two-layer model — the structure a slice fits into.
- A1 · Why an agile agent workflow