Unfolding AI: How ELA Educators can use AI to support Agentic Student Writing

Chris Proctor<br>
April 14, 2026

---

Assistant professor of Learning Sciences at UB

Founding director of UB's CS teacher preparation program

Former high school English teacher and middle school CS teacher
]

---

**My perspective**: How I'm thinking about AI in ELA
  - Epistemic tensions
  - Schools as laboratories
  - AI as a platform and a medium

**What's new?**
  - Dialogic identity writing with AI
  - Three authorial roles
  - AI affordances in Unfold Studio

**Demo**: Writing with AI in Unfold Studio

**What's next**: Declarative narrative
]

---

# My perspective: <br>Epistemic tensions

- When AI enters ELA, whose goals take priority? .cite[1]
- STEAM integration often asks "A" to serve "STEM" goals .cite[2]
- Computational thinking as a one-way bridge.cite[3]: ELA disciplines itself to computational goals, but computing does not discipline itself to ELA goals

---

# My perspective: <br>Schools as Laboratories for<br> New Literacies

Two models of literacy .cite[1, 2]:

- **Autonomous model**: literacy is a universal and inherently empowering skillset. It is our duty to give literacy to those who lack it. 
- **Ideological model**: literacies are specific social practices embedded in identites, communities, media, and power relations. 
  Literacies *can* be powerful in the right circumstances.

We don't know how AI will be powerful for today's youth, so why are we so fixated on telling them how to use it? Schools should be laboratories for discovering new practices.

---

# My perspective:<br>AI as a platform and a medium

Literacy practices at every scale are bound up with the technological infrastructure 
which enables them.cite[2]. AI is widespread and uniquely intimate.

ELA educators and researchers need to be **designing** AI interfaces—not just adopting them.

---

# A few contributions

---

# Writing with AI is dialogic

Writing is dialogic: meaning emerges through interaction among multiple voices; our words carry histories.cite[1]
Identity is a self one authors and occupies within a literacy practice .cite[2] Writer's workshop pedagogy cultivates authorial identity: students write as authors with something to say to their peers.cite[3]

Writing with AI changes the dialogic situation:
- AI models **culture and discourse**, not just language
- AI carries **language ideologies** around social position and genre .cite[4]
- Students' authorial identities become entangled with the voices of AI

.refs[
.anchor[1] Bakhtin, 1981
.anchor[2] Holland et al., 1998; Moje & Luke, 2009
.anchor[3] Dorn & Soffos, 2001
.anchor[4] Irvine & Gal, 2000; Proctor & Rish, 2025
]

---

# Three authorial roles

Three authorial roles .cite[1]:

- **Animator** — gives voice to words; inscribes them on the page
- **Author** — selects the sentiments and ideas
- **Principal** — is committed to; stands behind what the words say

These roles explain why AI feels acceptable in some contexts and fraudulent in others:

- AI-generated greeting card
- AI-generated eulogy
- AI-generated student essay

---

# Unfold Studio:<br>Couples therapy

---

# AI affordances in Unfold Studio

**Three AI affordances** each allow the author to delegate a literary element to AI, making collaboration with AI explicit and bounded:

- **`generate`**: Delegates sensory detail
- **`continue`**: Delegates plot
- **`agent`**: Delegates character

---

# `generate`: Delegating Detail

Generates text in response to a prompt; control returns to the story.

```
{generate("One sentence describing an alien ship's
long halls with three doors.")}
```

→ *"The hallway's endless metal panels gleamed under cold violet luminescence,
and three silent doors—one etched with shifting glyphs, one glowing with liquid
emerald light, and one veiled in crackling blue plasma—stand like sentinels..."*

---

# `continue`: Delegating Plot

Opens an unscripted but bounded exchange between reader and AI; AI guides toward a scripted target.

```
=== room1 ===
This room is filled with test tubes and you get stuck by some alien
goop on the ground. An Alien finds you in the room.
-> continue(-> ending5)

=== ending5 ===
The alien grabs you, puts you into one of the tubes, and leaves you
there for centuries.
-> END
```

---

# `agent`: Delegating Character

Requires a **target** (where the story resumes) and a **character** (a sub-story modeling the character's voice).

```
An alien scientist enters the room.
-> agent(-> after_encounter, -> alien_scientist)

=== alien_scientist ===
Cold. Methodical. Fascinated by human behavior.
"Why do they resist? The data is all that matters."
I have catalogued 47 species. Humans are... unpredictable.
-> END

=== after_encounter ===
The alien releases you and opens the door.
```

---

# Case Studies

---

# Study Context

- **Ten-week after-school writing club** at a high school (northeastern US)
- Structured as a **writer's workshop**
- Two focal cases selected as **contrasting examples** of dialogic authorial identity

**Data sources:**
- Sequence of story versions (every save produces a version)
- Records of authors' playthroughs
- Fieldnotes and ongoing discussions
- Pre- and post-surveys

---

# Case 1: Dragonbaby

- 9th-grade girl; collaborates with her 12th-grade sister
- Avid readers of fantasy; dedicated fan-fiction writers
- **No prior experience with generative AI**
- Story: *"Our actions have consequences"* — fan-fiction set in Percy Jackson

**Pattern of development:**
- **Early phase**: intense debugging + code edits (learning the language)
- **Middle phase**: intensive text and code editing
- **Later phase**: AI affordances layered onto existing story → **intensive playthroughs**
- After the workshop ended: still replaying the story weeks later

---

# Unfold Studio:<br>Worlds collide

---

# Dragonbaby: <br>Authorial Identity

Dragonbaby discovered the AI *knew her fandom*—Percy Jackson lore, characters, voice.

- **Animator** → frequently delegated to AI
- **Author** → increasingly *shared* with AI as co-author
- **Principal** → retained, but AI gained a subtle foothold

> Through intensive replaying and geeking out, Dragonbaby authored a **positive dialogic identity** embedded in AI-generated discourse—a fan-fiction literacy practice shared *with* the AI.

---

# Case 2: Quaezae

10th-grade girl; expressed moral opposition to generative AI
  - *"AI is ruining my friends' minds"*
  - Only used AI on friends' phones, not her own
Invited to experiment with AI in order to become better-informed in her opposition. She agreed and wrote a dystopian story: *"Tellers"*

> "AI has completely been enthralled into our lives... In this world, it's been completely accepted, or more accurately... **integrated**."

The distinction between *accepted* and *integrated* perfectly articulates the dialogic nature of AI.

---

# Unfold Studio:<br>Tellers

---

# Quaezae: Authorial Identity

- **Animator** → delegated to AI via `generate`
- **Author** → *not shared* with AI (wary, reluctant engagement)
- **Principal** → retained, in **adversarial** mode
- Asked for the reader's *name*, then injected it into AI prompts—creating the unsettling feeling of being put into dialogue with AI against your will

> Quaezae's authorial identity is also dialogically enmeshed with AI—but against her will. The story feels claustrophobic: an authorial voice rejecting something it recognizes as part of itself.

---

# What's next

Curriculum development and iteration. Join us!

---

Chris Proctor<br>
April 14, 2026