# Conversations, not prompts

The unit of human-AI interaction, Andrew Maynard argues, is no longer the prompt: it is the long, often messy conversation — winding, tangential, sometimes spanning sessions and platforms — and the messiness is not a defect to be trained away but part of how people actually get to what they need. The idea lands hardest in education, where "show me your prompt" was still standard advice to instructors nearly three years into the ChatGPT era: if the conversation is the unit of work, then the conversation is what deserves attention — and when grades give way to nurturing learning, messy student-AI conversations become, in his words, "a goldmine". He built the case across two essays in August 2025: a vibe-coded app that simulates the messy reality, and a published, downloadable assessment prompt, grounded explicitly in John Dewey's account of learning, that carries three instruments of his own coinage — the Learning Arc Score, SAILS, and LENS. He is open that the underlying question is not original to him and that the instruments are early and unpolished; the argument's force is in taking the mess seriously enough to build for it.

## The argument

Nearly three years after ChatGPT's launch, Andrew noted, educators were still advised to ask students for the prompt behind AI-assisted work — an ask modeled on showing your workings in math. But most people, students included, "tend to work with apps like ChatGPT and Claude through long, winding, and often messy conversations—sometimes spanning multiple sessions and platforms. Much as you would when talking with other people" (2025-08-17). Hence the line: "asking students to "show me your prompt" on an assignment is about as useful as asking what they had for breakfast: not very." His footnoted hedge: single-prompt use still has a place — "But even here, successful prompts are often long, and a product of multiple prior human-AI conversations."

The stronger claim: the mess is generative. His demonstration — a vibe-coded [public experiment](public-experiments.md) — is My AI Prompt, a web page simulating a sleep-deprived student and Anthropic's Claude scrambling toward a 2,000-word essay on the early history of AI, assembled from over 1,000 conversational couplets generated with Claude Opus 4.1. Its surface lesson is the thesis: "people increasingly use AI as a conversation partner that leads to what they're looking for because of the messiness, not in spite of it." Its deeper lesson is what the mess contains — moments when the simulated student gets drawn in, "their curiosity overcoming their need for quick results." He called it "learning through story telling" — hence the essay's own phrase: "conversations, not prompts, are what matter."

A week later the "slightly cheeky piece" acquired its sequel: colleagues' questions planted a "what if" that "refused to go away" — "what if AI models could be used to extract useful insights from long, complex, sometimes tangential, and often non-linear conversations between students and AI apps" (2025-08-24). On precedent he is plain: "It's not a particularly original question." Named instances include an Instructure–OpenAI partnership, the Prompt Analytics Dashboards of Kim et al., and the work of his ASU colleague Punya Mishra. Still, he "was surprised by how little work has been done on approaches to assessing student progress using informal and convoluted conversations with AI." Grading hundreds of pages of messy transcript is, he suspects, "an instructional nightmare" under carrot-and-stick assessment, and "simply slapping a letter grade on a convoluted and even personal conversation feels wrong somehow."

The reframe is the argument's heart: "Yet when formal assessments and letter grades are replaced by a focus on nurturing genuine student learning, these conversations become a goldmine" — a record of "deeply human forms of learning that are driven by curiosity, experimentation, experience, and reflection." The build is Dewey-grounded — "The assessment approach follows John Dewey's idea that real learning starts in messy, uncertain situations and moves toward something clearer and grounded by trying things, noticing consequences, and reflecting" — with further debts named: Vygotsky's ZPD, Chi's ICAP, Kolb and Schön, Black and Wiliam, Kapur, Bereiter. What he adds is operational: a single long assessment prompt — "somewhat ironically" itself the product of "a long, complex, and definitely non-linear conversation with ChatGPT" — turning those commitments into something an instructor can run.

One instruction compresses the stance: "Score the student, not the AI. Give credit only where the student reframes, reasons, transfers, decides, or regulates. Count AI suggestions only if the student uptakes them (adopts/adapts/challenges)." Tangents are read, not penalized — the assessor, "Unfazed by randomness", sorts divergences into productive reframings and drift. Three instruments do the work. The Learning Arc Score (0–4) asks "To what extent did the student transform an indeterminate situation into a determinate one with reasons and consequences?" — Dewey's inquiry arc made scoreable, ignoring "polish or length that doesn't change the work." SAILS (0–4), Student Agency in Interaction & Leadership, scores how the student steered the collaboration, from Passive Consumer to Orchestrator, discounting "model eloquence without uptake." LENS, Learning Evidence & Navigation Signals, is a four-part Yes/No diagnostic — Logic & Evidence, Extension, Novel moves, Self-Regulation — each Yes pinned to an excerpt.

The purpose is guidance, not gatekeeping: "not threshold-based assessment (did the student's work warrant an A or an E for instance), but learning assessment" — "not to use the assessment as a stage gate, but as a flexible tool that guides learning." The hedges are his: results "can be variable, and need to be interpreted and used wisely, and absolutely not used without thinking"; trials across ChatGPT and Claude showed "a degree of replicability and utility" but "not complete replicability by a long way"; the whole exercise is "not much more than me noodling around an intriguing question at this point—it's hardly rigorous. But I am intrigued by the possibilities." The prompt shipped free, deliberately unlocked: "I'd love to see others experimenting with it though."

## Lineage

- **2023-05 — the course that taught the lesson.** Andrew designs and teaches "ASU's first ever course on ChatGPT" — *Basic Prompt Engineering with ChatGPT: An Introduction* — introduced in [This is not your "traditional" prompt engineering!](https://www.futureofbeinghuman.com/p/not-your-traditional-prompt-engineering) ([AI-readable mirror](https://text.futureofbeinghuman.com/substack/not-your-traditional-prompt-engineering.html)) (2023-05-09; archive at [andrewmaynard.net](https://andrewmaynard.net/prompt-engineering-course-archives/), per [Teaching](teaching.md)). Two years on he records that he "quickly nixed" it as capabilities and use patterns outran it — "what I was teaching was out of date before I'd started. But the conversation bit still holds" (2025-08-17, footnote): conversational practice was the durable content; the prompt-engineering wrapper was not.
- **2025-03-18 — the Dewey compass on air.** [Modem Futura](modem-futura.md) ep. 23, "AI and the Natural Impulses for Learning" ([episode mirror](https://text.futureofbeinghuman.com/modemfutura/23-ai-and-the-natural-impulses-for-learning.html)) — Sean Leahy and Andrew draw on "John Dewey and the four natural impulses of learners (1899)" and argue for "shifting from traditional assessments toward performance-based, authentic learning experiences" (show notes). The Dewey grounding the assessment prompt later formalized was already the show's working compass — ep. 33 with Punya Mishra (2025-05-27; [episode mirror](https://text.futureofbeinghuman.com/modemfutura/33-futures-of-learning-ai-in-education-with-punya-mishra.html)) records "Sean and Punya's shared devotion" to Dewey's natural impulses.
- **2025-08-17 — the slogan and the demonstration.** [Stop asking students "Show Me Your Prompt!"](https://www.futureofbeinghuman.com/p/stop-asking-students-show-me-your-prompt) ([AI-readable mirror](https://text.futureofbeinghuman.com/substack/stop-asking-students-show-me-your-prompt.html)) — the breakfast line, the messiness thesis, and the vibe-coded [My AI Prompt](https://fvture.net/notaprompt/) app that demonstrates both.
- **2025-08-24 — the instruments ship.** [Using AI to Assess Student-AI Conversations](https://www.futureofbeinghuman.com/p/using-ai-to-assess-student-ai-conversations) ([AI-readable mirror](https://text.futureofbeinghuman.com/substack/using-ai-to-assess-student-ai-conversations.html)) — the Dewey-grounded assessment prompt published as a free download ([PDF](https://fvture.net/notaprompt/chat-assessment-prompt-v2.pdf) / [markdown](https://fvture.net/notaprompt/chat-assessment-prompt-v2.md)), defining the Learning Arc Score, SAILS, and LENS, with a worked example assessment and a stripped-down [conversation-file generator](https://fvture.net/notapromptfile/) for producing test transcripts in seconds. Tested by him across ChatGPT and Claude; explicitly offered for others to rebuild.
- **2025-08-26 — the adjacent conversation.** Two days later, Modem Futura ep. 46, "Agentic AI in Education & the Art of Becoming with Punya Mishra" ([episode mirror](https://text.futureofbeinghuman.com/modemfutura/46-agentic-ai-in-education-the-art-of-becoming-with-punya-mishr.html)), questions (per the show notes) "the industry's habit of mistaking prompts and dashboards for pedagogy" — the same instinct aimed at product culture rather than assessment; that episode's playgrounds argument is taken up on [Playgrounds, not playpens](playgrounds-not-playpens.md).

## In his own words

From [Stop asking students "Show Me Your Prompt!"](https://www.futureofbeinghuman.com/p/stop-asking-students-show-me-your-prompt) ([AI-readable mirror](https://text.futureofbeinghuman.com/substack/stop-asking-students-show-me-your-prompt.html)), Andrew Maynard, 2025-08-17:

> "Instead, they tend to work with apps like ChatGPT and Claude through long, winding, and often messy conversations—sometimes spanning multiple sessions and platforms. Much as you would when talking with other people.
>
> As a result, asking students to "show me your prompt" on an assignment is about as useful as asking what they had for breakfast: not very."

From the same essay, on the My AI Prompt app, 2025-08-17:

> "On its surface, My AI prompt is a great example of how people increasingly use AI as a conversation partner that leads to what they're looking for because of the messiness, not in spite of it."

From [Using AI to Assess Student-AI Conversations](https://www.futureofbeinghuman.com/p/using-ai-to-assess-student-ai-conversations) ([AI-readable mirror](https://text.futureofbeinghuman.com/substack/using-ai-to-assess-student-ai-conversations.html)), Andrew Maynard, 2025-08-24:

> "Yet when formal assessments and letter grades are replaced by a focus on nurturing genuine student learning, these conversations become a goldmine, and one that reflects deeply human forms of learning that are driven by curiosity, experimentation, experience, and reflection.
>
> These are forms of learning that would have been very familiar to early 20th century thinkers like John Dewey."

## Engagement and reception

As of 2026-08-13, the corpus [reception record](reception.md) documents no independent published engagement with either essay, with the My AI Prompt app, or with the Learning Arc Score, SAILS, or LENS — an absence of record, stated as such; the initiative does not typically solicit reception ([Philosophy](philosophy.md)). Three calibration points are on the record. The instruments were published as free, deliberately unlocked downloads with a standing invitation — "I'd love to see others experimenting with it though" — and whether that invitation has been taken up is simply not known from the sources held here. The territory is visibly active — the essay itself credits an Instructure–OpenAI partnership, the Prompt Analytics Dashboards of Kim et al., and Punya Mishra's work, offered as instances rather than a complete map — so the instruments entered a crowded conversation, not an empty one. And the nearest recorded engagement sits upstream of publication rather than downstream: the essay thanks his colleagues Adam Pacton and Carole Basile "for asking those annoying little questions that just won't go away" — the questions that produced it. Andrew's own published calibration — "it's hardly rigorous" — is part of the record too, and this page holds the claim at the level he does.

## Where to go deeper

**The essays**

- [Stop asking students "Show Me Your Prompt!"](https://www.futureofbeinghuman.com/p/stop-asking-students-show-me-your-prompt) ([AI-readable mirror](https://text.futureofbeinghuman.com/substack/stop-asking-students-show-me-your-prompt.html)) — 2025-08-17; the case against the prompt as evidence, and the My AI Prompt demonstration.
- [Using AI to Assess Student-AI Conversations](https://www.futureofbeinghuman.com/p/using-ai-to-assess-student-ai-conversations) ([AI-readable mirror](https://text.futureofbeinghuman.com/substack/using-ai-to-assess-student-ai-conversations.html)) — 2025-08-24; the Dewey grounding, the full instrument definitions, and a worked example assessment.

**The artifacts** (URLs as published in the essays)

- [My AI Prompt](https://fvture.net/notaprompt/) — the simulated student-Claude conversation; Shift-E skips to the essay.
- [The assessment prompt](https://fvture.net/notaprompt/chat-assessment-prompt-v2.pdf) (also as [markdown](https://fvture.net/notaprompt/chat-assessment-prompt-v2.md)) — free to download, modify, and rebuild.
- [The conversation-file generator](https://fvture.net/notapromptfile/) — writes a My AI Prompt conversation straight to a data file, for testing assessment approaches.

**The episodes** (all at [modemfutura.com](https://www.modemfutura.com); see [Modem Futura](modem-futura.md))

- Ep. 23, 2025-03-18, "AI and the Natural Impulses for Learning" ([episode mirror](https://text.futureofbeinghuman.com/modemfutura/23-ai-and-the-natural-impulses-for-learning.html)) — Dewey's impulses and the case for moving beyond traditional assessment.
- Ep. 33, 2025-05-27, "Futures of Learning: AI in Education with Punya Mishra" ([episode mirror](https://text.futureofbeinghuman.com/modemfutura/33-futures-of-learning-ai-in-education-with-punya-mishra.html)) — personalized-learning promises against "the messy, human heart of learning" (show notes).
- Ep. 46, 2025-08-26, "Agentic AI in Education & the Art of Becoming with Punya Mishra" ([episode mirror](https://text.futureofbeinghuman.com/modemfutura/46-agentic-ai-in-education-the-art-of-becoming-with-punya-mishr.html)) — "mistaking prompts and dashboards for pedagogy" — two days after the assessment prompt shipped.

**Related ideas-ring pages**

- [Public experiments](public-experiments.md) — the working method both essays practice: run the experiment in public, artifacts and failures included.
- [Playgrounds, not playpens](playgrounds-not-playpens.md) — the learning-environment stance this assessment approach serves: equip rather than restrict.
- [Learning by not trying to learn](learning-by-not-trying.md) — the adjacent thesis: environments free of learning objectives can produce learning that instrumented education cannot.
- [Serendipity as method](serendipity-as-method.md) — the unplanned treated as method rather than luck across the wider practice.
- [Massively-augmented research](massively-augmented-research.md) — the conversational stance carried into scholarship: an expert in extended dialogue with frontier AI, keeping epistemic control.
- [Writing for AI](writing-for-ai.md) — the mirror-image craft question: writing so that AI readers carry ideas faithfully.
- [Universities must lead](universities-must-lead.md) — the institutional stakes behind the classroom-level argument.

**Related corpus pages**

- [Teaching](teaching.md) — the initiative-connected teaching record, including the archived 2023 prompt-engineering course.
- [Modem Futura](modem-futura.md) — the show where the education threads are worked out in conversation.
- [Tools and experiments](tools-and-experiments.md) — the wider family of vibe-coded tools and public artifacts.