Instructional Design for AI Agents

Instructional designers have spent decades learning how to transfer a skill to a new learner. Most agent prompts ignore all of it. This guide applies learning objectives, task analysis, scaffolding, worked examples, and assessment to the way agents are instructed, so that behavior is designed, taught, and measured instead of patched.

What you will be able to do

• Write behavioral objectives for an agent that can be tested.
• Break a job into tasks and decide which to teach, tool, or forbid.
• Use worked examples and scaffolding instead of rule lists.
• Assess agent behavior with the same rigor as a learning outcome.

Inside the guide

1. 01

   Prompts are curricula

   What instructional design already knows about teaching a task.

2. 02

   Objectives and task analysis

   Defining the job before writing a word of instruction.

3. 03

   Scaffolding and examples

   Worked examples, procedures, and progressive disclosure.

4. 04

   Assessment

   Evaluating behavior against objectives, not vibes.

5. 05

   Maintenance

   Updating instruction as the job and the model change.

From the guide

A rule list is the weakest form of instruction for people, and it is the weakest form for models too. A worked example of the right behavior in context teaches more than ten prohibitions.

Who it is for

• Teams responsible for agent prompts, policies, and procedures.
• Instructional designers and educators moving into AI work.
• Product owners who need agents to behave consistently across releases.