Sketch Your First Agent on Paper

Sketch your agent on paper before you write any code, and you'll spot exactly where it's likely to break. This lesson walks through a rough diagram for the Study Helper Agent.

Let's think about this for a second

A paper agent has no UI yet. You're just drawing boxes: where input comes in, where output goes out, which tools get used, where it stops, and where it asks a human. Skip this step and jump straight to code, and your tools, prompts, and error handling all get tangled together — hard to untangle later. Same idea as writing a spec before vibe coding: with agents, you sketch the architecture first.

Let's connect this to everyday life

Study Helper paper sketch — User question → Goal checker → list_notes / read_note → draft_summary → Human preview → (optional) save draft. Next to it, write out the failure paths. What does it say if the file doesn't exist? If a tool errors out, does it retry or stop? If there's no answer, does it say "not found in the notes" or does it guess? Decide right here, at this stage: no guessing, ever.

Diagram the input, tools, output, and stop point
Write out the paths for two error cases
Note down: when unsure, never guess

Let's try it out together

text

User question
   ↓
Goal checker (is this a study question?)
   ↓
list_notes → read_note
   ↓
draft_summary (3 bullets)
   ↓
Preview to human
   ↓
STOP  or  save only after approval

If file missing → say so and stop
If unsure → say "မှတ်စုထဲမတွေ့" and stop

You should see

You'll be able to explain the Study Helper Agent's diagram to someone else.

5-minute try

On paper or in a file, sketch your own agent diagram. Keep it to 6 boxes or fewer. Include 2 error paths.

A quick word of caution

If any box in your diagram is basically "magic," you're not ready to code yet. Break that box down into a tool or a rule.