5 minModule 1 — The AI Development Problem

L1.2: The Three Pillars

The Problem

You have two runs and two notes from L1.1. The outputs were unpredictable. The mismatches were real. But "it was bad" is not actionable — you need to know why it failed and what specifically would fix it. Without that mapping, you'll keep vibing and hoping, just with more awareness.

The Solution

Every failure from the L1.1 runs falls into one of three categories, and each category maps directly to a practice that fixes it:

Failure category	Root cause	Practice that fixes it
Requirements gap	The prompt lacked specificity; Copilot guessed intent	Instructions — tell Copilot what the project is, what stack to use, what to never do
Technical mismatch	The generated code was inconsistent, wrong stack, or mismatched structure	Agents — create role-specific agents with constrained context
Quality/risk issue	Missing tests, error handling, security gaps, or brittle architecture	Specs — define acceptance criteria and task scope before generating code

These are the three pillars:

1. Instructions — set the context and boundaries Copilot always operates within
2. Agents — create specialized roles for specific, repeatable tasks
3. Specs — define what done means before writing a line of code

Note: A fourth practice — Skills/Prompt files — extends the workflow with reusable automated sequences. It is not taught as a dedicated session but is introduced briefly in L5.3.

Treating every Copilot failure as a prompt problem and trying to fix it by rephrasing. This is pattern-matching on symptoms. The root cause is missing structure — and rephrasing doesn't add structure.

Categorize the failure, identify the root cause, and map it to the pillar that addresses it. Then build that pillar so the failure can't recur.

Exercise

Use your two follow-up notes from L1.1.

Step 1 — Identify failure signals

Review your Run 1 and Run 2 notes. Identify a minimum of 3 failure signals — specific things that went wrong, were missing, or were unexpected.

• Expected result: a list of at least 3 concrete observations pulled from your notes.
• Why this matters: generic complaints ("it was wrong") don't drive improvement; specific observations ("it generated React but I needed vanilla TypeScript") do.

Requirement: You must include at least one signal from each of the following categories:

• Requirements gap
• Technical mismatch
• Quality/risk issue

Step 2 — Write your reflection

For each failure signal, write one bullet entry using this format:

- Category: [Requirements gap | Technical mismatch | Quality/risk issue]
  Observation: [What you noticed in the output]
  Why it happened: [Root cause — what was missing from the prompt or context]
  Which practice fixes it: [Instructions | Agents | Specs]

Requirement: Each signal maps to exactly one practice.

Example:

- Category: Requirements gap
  Observation: Copilot generated a React app, but I didn't specify a framework
  Why it happened: The prompt said "study planner" with no tech stack context
  Which practice fixes it: Instructions — add a workspace instructions file specifying the stack

• Expected result: 3 or more reflection bullets, each with all four fields filled.
• Why this matters: this reflection is the plan for Module 2. Each "which practice fixes it" entry is a thing you will build.

Step 3 — Preview what's next

Look at your reflection bullets. Count how many map to each pillar:

• Instructions → you'll build those in L2.1

• Agents → you'll build those in L2.2

• Specs → you'll build those in M3

• Expected result: you have at least one entry pointing to each module.

• Why this matters: the rest of the workshop is designed to fix exactly what you just documented.

Checkpoint

• [ ] I identified at least 3 failure signals from my L1.1 runs
• [ ] I included at least one signal from each category: requirements gap, technical mismatch, quality/risk issue
• [ ] I wrote a reflection entry for each signal with all four fields: Category, Observation, Why it happened, Which practice fixes it
• [ ] I can name which module fixes each of my recorded failures

---

← Previous Lesson: L1.1 The Vibing Trap · Next Module: M2 Copilot Best Practices →