AI Decision Lab

Step 1

Choose the problem before the technology

Select a case. First ask whether an LLM belongs in the solution at all.

Step 2

Complete the implementation canvas

The canvas remains the main planning artifact. Use this page to connect it to the Lecture 2 experiments.

AI Implementation Canvas

Map the current process, AI role, data and context, human boundary, evidence, value, and smallest credible pilot.

Open the canvas ↗

Canvas link placeholder

Canvas checkpoint Confirm before moving on

We defined the current workflow and baseline. We stated what someone will do differently. We selected the necessary data and context. We defined the human approval boundary. We proposed the smallest credible pilot. We considered a non-LLM alternative.

Step 3

Turn the idea into a precise task

A prompt is not a clever sentence. It is a compact task and context specification.

Goal What must be produced? User Who acts on the output? Relevant context Include only what is needed Instructions Ordered actions Constraints What must it never do? Evidence rule How are claims supported? Output format Structure over fluent prose Abstention and escalation When should it stop?

Working prompt Edit before testing

Run the prompt in a model outside this page. Do not use confidential or personal data in classroom tests.

Step 4

Try to make the system fail

Repeat a run when useful. Record failures, uncertainty and abstention—not just the best-looking output.

Optional model comparison Repeat one test with another model

Models or configurations Preferred option and why Trade-off observed

Step 5

Decide—and make the decision reversible

Your recommendation should follow from the evidence, not from enthusiasm for AI.

Decision confidence 50%

Why? Strongest evidence Greatest remaining risk What could reverse the decision?

Implementation roadmap

Start in class; complete and justify it at home.

Phase	What will we do?	Evidence required to continue