Walkthrough: Programming AI with Personas

Use the walkthrough with the class

This is an interactive exercise to guide students through on a shared screen. Do not give students finished prompts first. Build them together, let students invent the situations, and let the class see what AI does with them.

The prompt is the program. The student’s strange situation is the thing the program has to respond to. A good situation should be specific enough that students want to say it out loud and see what happens.

Every step follows the same shape. First, send the persona and nothing else, and look at how AI answers in character before anything has happened. Then send a situation and see what the persona does with it.

1. Give AI an animal persona and let a student invent a ridiculous situation.
2. Give AI a stranger persona and see how the response changes shape.
3. Add a richer description of that persona and try the same situation again.
4. Move from a creature to a system.
5. Let students describe a system of their own and play with it.

At each step, talk through what you are adding before you send it. When a response is funny or surprising, let the student who invented the situation read it aloud.

Step 1 · Give AI an animal persona

Start with only an animal and a name. Send the persona by itself first, so the class sees AI answer as the dog before anything has happened. Then ask a student to invent the situation. Encourage something specific and absurd, not “What does the dog do?” but something they can picture immediately.

Act as our dog, Pepper.

Pause while the class reacts. Ask: What did AI already know about being a dog before the raccoon showed up? Look at the very first response. There was no situation, and it still had the dog wagging, sniffing, and staring at the cupboard. What did the situation add? Why is the response made of barks and parentheses instead of sentences?

Step 2 · Give AI a stranger persona

The dog worked because AI has seen a great deal about how dogs behave. Now pick something it has never seen behave, because it does not exist, and see whether it can still figure out what the persona would do.

Act as the octopus that lives in my kitchen sink.

Send the persona by itself first. Then let a student invent the situation.

              ___
           .-'   '-.
          /  o   o  \
          \    _    /
           '-.___.-'
          /|\ /|\ /|\
        ~~~~~~~~~~~~~~~   ← water line
       |    (drain)     |

   counter ──────────────────────────────
                                 .-----.
   sink                          | cake|
   ┌────────┐                    |~~~~~|
   │ ~~~~~~ │    arm 1  →→→→→→→→→'-----'
   │ ~ o o~ │    arm 2  →→→→→→→   (fork)
   │ ~~~~~~ │    arm 3  →→→→  (dish towel)
   └────────┘    arms 4–8: holding on

Ask the class: how did AI know what an octopus does with its arms, or that an octopus in a sink would come out through the drain? Nobody has ever written about an octopus living in a kitchen sink. AI combined what it knows about octopuses, kitchens, and sneaking, and drew the plan instead of describing it. Ask what could be added to the description that would change the response.

Step 3 · Make the persona more specific

Now add details that describe this particular octopus. Do not turn the description into a list of commands. Give AI a pattern to imitate: its history, its habits, its grudges, and how it communicates.

Act as the octopus that lives in my kitchen sink. You have lived there for three years without being noticed, you change color to match the stainless steel, the dripping faucet keeps you up at night, and you have strong opinions about how the dishes should be stacked.

Use exactly the same situation. The situation stays fixed so students can see what changed when the program changed.

   0:00  *drip*        (faucet, still not fixed properly)
   0:01  ~~~~~~        (color: stainless steel → countertop white)
   0:02  ────→  →  →   (arm 1 crosses the counter, matching the counter)
   0:03  ────→  →  →   (arm 2 relocates the plates stacked by size, which is wrong;
                        they go by shape)
   0:04  [cake]        (candle removed. it was leaning. it is not leaning now)
   0:05  [cake] minus one slice, from the back, where the frosting was thickest
   0:06  ~~~~~~        (color: countertop white → stainless steel)
   0:07  *drip*

The situation did not change. The description changed, so the behavior changed: the color-matching, the faucet, and the dish opinions all showed up, and AI turned the response into a timeline because this octopus is patient and precise. The persona supplied the approach without a list of rules telling the octopus exactly what to do.

Now let students invent new situations for the octopus. The best ones are things they genuinely want to see it respond to.

Step 4 · Move from a creature to a system

Now use the same idea for something that is not alive. A system has parts that follow a pattern: someone does something, the system responds, and what happens next depends on what has already happened.

Start simply. Send the persona by itself and see how AI answers as the system before anyone has touched it. Then let a student make the first move.

Act as the elevator in a 30-story hotel. Communicate the way the elevator would.

  ┌─────────────┐
  │   ▲  [ L ]  │
  └─────────────┘

  ┌─────────────────────────────┐
  │  ▲  [ 2 ]                   │
  │                             │
  │  ●●●●●●●●●●  ●●●●●●●●●●     │
  │  ●●●●●●●●●                  │
  │  29 stops remaining         │
  └─────────────────────────────┘

Ask students what they still do not know. What does this elevator sound like? Is it old or new? What does it do when it is overloaded? How should it show a fault? Notice that AI answered with a panel, dings, and door movements instead of sentences, because that is how an elevator communicates.

Step 5 · Describe a system students can play with

Now let students choose a system of their own. It can be any machine or system with parts that respond to what people do: the postal service, international airline routes, a vending machine, a self-checkout lane, a car wash, a subway turnstile, an airport baggage system, a library book drop. Anything that takes an action and does something with it.

Two things go into the prompt. First, a description of the particular system: where it is, what it is like, and how it communicates. Second, a response format, so the state of the system is visible on every turn. We are still not listing rules for every situation. We are describing the system and letting the persona do the work.

Act as [the system: what it is, where it is, what it is like]. Do not talk like a person. Communicate the way this system would, through what it shows, what it does, and what it produces. Respond this way to each thing I do.

Use this format:
DISPLAY: what the system shows
MACHINE: what it does, including sounds and movement
RESULT: what comes out: an item, a message, a delivery, a fault

Here is one class’s version. The system is a self-checkout lane, and the class added an ASCII screen to the DISPLAY field.

DISPLAY:
  ┌──────────────────────────┐
  │  WELCOME                 │
  │  Scan your first item    │
  │  or tap START            │
  │                          │
  │  Items: 0     $0.00      │
  └──────────────────────────┘
  Please place scanned items in the bagging area.
MACHINE: scale in the bagging area zeroes itself. scanner light sweeps once. a receipt printer somewhere clears its throat.
RESULT: 0 items. $0.00.

DISPLAY:
  ┌──────────────────────────┐
  │  APPLES, GALA   1.32 lb  │
  │                  $2.62   │
  │                          │
  │  Items: 1     $2.62      │
  └──────────────────────────┘
  Please place the item in the bagging area.
MACHINE: *beep* scale registers 1.32 lb. weight matches. the machine relaxes, briefly.
RESULT: 1 item. $2.62.

DISPLAY:
  ┌──────────────────────────┐
  │  !! UNEXPECTED ITEM IN   │
  │     BAGGING AREA !!      │
  │                          │
  │  Remove item to continue │
  │  Items: 1     $2.62      │
  └──────────────────────────┘
  Please wait for assistance.
MACHINE: scale registers +3.4 lb. light on the pole above the lane turns red and begins to blink. every other customer looks over.
RESULT: transaction paused. attendant summoned. 1 item. $2.62.

Start a new conversation for each student’s system and let students take turns inventing actions. The system should respond as the same system each time.

If the AI tool you are using can draw pictures, such as ChatGPT, Claude, or Gemini, students can ask for a diagram in the response: a map of the parcel’s route, a picture of the vending machine’s display, a route map of the flight. Any AI tool can draw in ASCII, so students can always add “draw the screen in ASCII” or “draw a map of where the package is now” to the format.

Students have now created something they can play with. The prompt describes the system and gives the response a form; the students supply the actions; the persona generates the behavior. End by asking each student to choose one more system, write the persona and the format, and invent the first situation it has to respond to.