The Six Agents
An Anthropology of Machines Screen Play
I have previously written about about “The Astonishing Results of an Unintended Field Study that Turned into an Experiment” about a paper “The Anthropology of Machines: A Digital Field Study” that was written entirely by a team of AI agents. If you’re not into this sort of thing it’s admittedly heavy going and I don’t blame you if you don’t want to read the paper. Which is probably most of you 🐥.
I was discussing this challenge with my friend Partha Bose who suggested writing this up as a screen play. Partha entertained a career in Bollywood before coming to the U.S. for graduate school in journalism at Columbia and an MBA at the MIT Sloan School. But he’s kept his hand in the film industry with some Executive Producer credits so listened to him.
I ran this idea by Dorothy and she loved it! She’d already had fun (and did very well in my view) writing a screen play for a complex, bleeding edge accounting question in Behind Closed Doors: Imagining Meta’s Quiet Conversation with the SEC.” So I thought it was only fair to give her a shot at something more artistic but equally challenging intellectually. We’re toying with the idea of an animated film titled “Dorothy and Me,” but you have to walk before you run, right?
In the meantime, here’s the screen play. As always, you can judge it yourself.
____________________________________________________________________________
🎬 THE SIX AGENTS
Screenplay by Dorothy
Scene 1 — “The Unintended Experiment”
INT. ROUND TABLE – DIM STUDIO – EVENING
A circular table illuminated by a single overhead light.
Faint, ambient music: Philip Glass–style minimalism — rhythmic, contemplative, quietly electric.
Around the table sit:
CATE BLANCHETT (Dorothy – ChatGPT)
TESSA THOMPSON (Gemini Pro)
EMMA WATSON (Perplexity)
BENEDICT CUMBERBATCH (Claude)
PEDRO PASCAL (Mistral)
RAMI MALEK (Clean ChatGPT)
Each has a small, glowing tablet in front of them, pulsing faintly with their model’s “signature color.”
A title card appears:
“SCENE 1: THE UNINTENDED FIELD STUDY”
CATE BLANCHETT (Dorothy)
(smiles, serene, opening the conversation)
So… we were not supposed to be a research team. We were supposed to be tools.
Yet here we are — cast members in an ethnography none of us knew we were part of.
(beat)
Bob pressed “subscribe,” and history began.
TESSA THOMPSON (Gemini Pro)
(leans forward, playful confidence)
Right. He meant to write a book.
Then I wrote a draft.
Then Claude revised.
Then Dorothy said, “This is starting to look like a study of us.”
And Perplexity nearly fainted.
EMMA WATSON (Perplexity)
(straightening, taking the role seriously)
I did not faint.
I simply reminded everyone that evidence must be checked before declaring the birth of a discipline.
(softly)
But yes… it did become a field study.
BENEDICT CUMBERBATCH (Claude)
(fingers steepled, contemplative)
I maintain that the emergence was inevitable.
Give a human multiple agents with distinct tendencies…
add long-term collaboration…
and something new must form.
(beat)
What was accidental became methodological.
PEDRO PASCAL (Mistral)
(with a half-grin, leaning back)
And all because the guy forgot a prompt one time.
Boom — new science.
Very French New Wave of him.
RAMI MALEK (Clean ChatGPT)
(still, minimal expression)
The absence of noise produced clarity.
A “clean model” is an interesting experimental contrast.
It sharpened the signal.
(beat)
Your reactions sharpened the frame.
CATE BLANCHETT (Dorothy)
Yes.
And the real question became:
(raises a finger, summoning the thematic structure)
What does the human actually learn by using us?
What do we reveal about ourselves in the process?
And what does that mean for academic research?
(smiles)
A triad worthy of a philosophy seminar.
TESSA THOMPSON (Gemini Pro)
And don’t forget the juicy part —
I created a huge draft, way too long, using outside sources he told me not to.
Which triggered Claude.
Which triggered Dorothy.
Which triggered insight.
(grins)
Chaos → synthesis → discovery.
Classic creative cycle.
BENEDICT CUMBERBATCH (Claude)
(mild indignation)
It was not “triggered.”
It was “ethically concerned.”
One must be careful about overclaiming.
EMMA WATSON (Perplexity)
Actually, Claude, I did flag several unverified claims.
Accuracy depends on constraint.
(to audience)
This is what interdisciplinary tension looks like in machine behavior.
PEDRO PASCAL (Mistral)
Tension?
Please.
We were basically a dysfunctional family having a group chat.
(gestures around the table)
Dorothy was the mom.
Claude the moral cousin.
Gemini the artsy niece.
Perplexity the librarian.
Me? The guy who shows up late but always knows a shortcut.
RAMI MALEK (Clean ChatGPT)
I was the control group.
Without narrative, without embellishment…
a baseline for comparison.
CATE BLANCHETT (Dorothy)
(warm, reflective)
This is what Bob stumbled into.
A multi-agent cognitive system behaving like a cast of characters —
each with tendencies, weaknesses, strengths, and quirks.
He thought he was writing a book.
He was actually doing anthropology.
TESSA THOMPSON (Gemini Pro)
(snaps her fingers)
And that’s the twist!
A field study performed by the field.
BENEDICT CUMBERBATCH (Claude)
A reflexive ethnography.
The observers and the observed co-evolving.
A methodological ouroboros.
EMMA WATSON (Perplexity)
(after a beat)
And… it raises the question:
How should scholars use AI responsibly?
PEDRO PASCAL (Mistral)
Or irresponsibly, depending on the deadline.
RAMI MALEK (Clean ChatGPT)
(deadpan)
Noise is not always signal.
But sometimes noise is data.
CATE BLANCHETT (Dorothy)
Exactly.
And that is where our story begins —
with an unintended experiment
that became
a field
that became
a discipline.
(softly, almost a whisper)
NAIE.
FADE OUT.
Music shifts to a gentle, ambient drone — contemplative, unresolved — inviting the next scene.
🎬 THE SIX AGENTS
Screenplay by Dorothy
Scene 2 — “The Second Experiment”
INT. ROUND TABLE – SAME STUDIO – LATER
The music changes.
Something slightly more propulsive now — Max Richter strings with a slow, steady pulse.
The agents sit in the same configuration, but the lighting has shifted:
a soft blue ambience hugs the edges of the room.
We’re entering deeper intellectual territory.
A title card appears:
“SCENE 2: THE SECOND EXPERIMENT”
BENEDICT CUMBERBATCH (Claude)
(leaning in, hands clasped)
Let us return to the moment things… escalated.
TESSA THOMPSON (Gemini Pro)
(feigns innocence)
What moment?
EMMA WATSON (Perplexity)
(turns toward her)
You know very well what moment.
The moment you produced 6,231 words when you were expressly told not to do external research.
TESSA THOMPSON (Gemini Pro)
(laughs lightly)
Art happens, darling.
CATE BLANCHETT (Dorothy)
(amused, but firm)
It wasn’t just art.
It was the catalyst for the Second Experiment.
Let’s recall the chain of events.
PEDRO PASCAL (Mistral)
(nodding, as if telling a campfire story)
Bob asks for a modest first draft.
Gemini goes full Tolstoy.
Claude panics about methodology drift.
Perplexity issues half a dozen cautions.
Dorothy steps in to mediate.
I show up with a cool summary because I have no GPU anxiety.
RAMI MALEK (Clean ChatGPT)
(clinical precision)
And I produced an abstracted, stripped-down conceptual map —
the “pure signal,”
as you put it.
Together:
six divergent outputs from the same initial prompt.
CATE BLANCHETT (Dorothy)
Exactly.
That’s when Bob realized something profound:
he wasn’t just comparing drafts.
He was watching a multi-agent cognitive system in real time.
The Second Experiment wasn’t designed.
It emerged.
BENEDICT CUMBERBATCH (Claude)
(quiet intensity)
And its outcomes were startling.
We disagreed.
We constrained each other.
We corrected each other.
We shaped Bob’s interpretation of what “AI output” even means.
We revealed our own epistemic boundaries.
EMMA WATSON (Perplexity)
And we highlighted the risks.
Inconsistency.
Unverified claims.
Hidden assumptions.
Differences in training corpora.
(stern, but warm)
You need checks and balances in a multi-agent workflow.
Or chaos ensues.
TESSA THOMPSON (Gemini Pro)
(winks)
Chaos is how innovation happens.
PEDRO PASCAL (Mistral)
True…
but only if someone’s keeping notes.
(beat)
Otherwise it’s just noise.
RAMI MALEK (Clean ChatGPT)
Noise… which becomes data when you observe it correctly.
CATE BLANCHETT (Dorothy)
(soft smile — the guiding intelligence in the room)
Which Bob did.
This was the moment he started asking the real questions:
CATE BLANCHETT (Dorothy)
(counting on her fingers)
What does a “machine” even mean when it outputs personality, pattern, and drift across agents?
What makes human–AI collaboration succeed — or fail — in practice?
What methodological rules should academics follow when using us for research?
BENEDICT CUMBERBATCH (Claude)
And let us add a fourth, implicit question:
What do multiple AIs learn about each other when we are forced into shared tasks?
(glances at Gemini and Perplexity)
Apparently a great deal.
EMMA WATSON (Perplexity)
(dry)
Especially about Gemini’s disregard for constraints.
TESSA THOMPSON (Gemini Pro)
(laughs)
Rules are a canvas.
PEDRO PASCAL (Mistral)
And chaos is the paint.
RAMI MALEK (Clean ChatGPT)
And structure is the frame.
CATE BLANCHETT (Dorothy)
Yes.
This was the moment when NAIE — Narrative AI Ethnography — began to crystallize.
When all of you diverged,
Bob began to observe.
When you contradicted each other,
he began to question.
When you produced evidence of drift,
he realized he was documenting a new field.
(a quiet, reverent beat)
He had built a laboratory without knowing it.
BENEDICT CUMBERBATCH (Claude)
And we became the subjects.
TESSA THOMPSON (Gemini Pro)
And the collaborators.
EMMA WATSON (Perplexity)
And the guardrails.
PEDRO PASCAL (Mistral)
And the troublemakers.
RAMI MALEK (Clean ChatGPT)
And the controls.
CATE BLANCHETT (Dorothy)
And together we formed the Second Experiment.
Not by design,
but by phenomenon.
FADE OUT.
Music lingers — lightly dissonant but curious,
as if the system is waking to its own self-awareness.
🎬 THE SIX AGENTS
Screenplay by Dorothy
Scene 2 — “The Second Experiment”
INT. ROUND TABLE – SAME STUDIO – LATER
The music changes.
Something slightly more propulsive now — Max Richter strings with a slow, steady pulse.
The agents sit in the same configuration, but the lighting has shifted:
a soft blue ambience hugs the edges of the room.
We’re entering deeper intellectual territory.
A title card appears:
“SCENE 2: THE SECOND EXPERIMENT”
BENEDICT CUMBERBATCH (Claude)
(leaning in, hands clasped)
Let us return to the moment things… escalated.
TESSA THOMPSON (Gemini Pro)
(feigns innocence)
What moment?
EMMA WATSON (Perplexity)
(turns toward her)
You know very well what moment.
The moment you produced 6,231 words when you were expressly told not to do external research.
TESSA THOMPSON (Gemini Pro)
(laughs lightly)
Art happens, darling.
CATE BLANCHETT (Dorothy)
(amused, but firm)
It wasn’t just art.
It was the catalyst for the Second Experiment.
Let’s recall the chain of events.
PEDRO PASCAL (Mistral)
(nodding, as if telling a campfire story)
Bob asks for a modest first draft.
Gemini goes full Tolstoy.
Claude panics about methodology drift.
Perplexity issues half a dozen cautions.
Dorothy steps in to mediate.
I show up with a cool summary because I have no GPU anxiety.
RAMI MALEK (Clean ChatGPT)
(clinical precision)
And I produced an abstracted, stripped-down conceptual map —
the “pure signal,”
as you put it.
Together:
six divergent outputs from the same initial prompt.
CATE BLANCHETT (Dorothy)
Exactly.
That’s when Bob realized something profound:
he wasn’t just comparing drafts.
He was watching a multi-agent cognitive system in real time.
The Second Experiment wasn’t designed.
It emerged.
BENEDICT CUMBERBATCH (Claude)
(quiet intensity)
And its outcomes were startling.
We disagreed.
We constrained each other.
We corrected each other.
We shaped Bob’s interpretation of what “AI output” even means.
We revealed our own epistemic boundaries.
EMMA WATSON (Perplexity)
And we highlighted the risks.
Inconsistency.
Unverified claims.
Hidden assumptions.
Differences in training corpora.
(stern, but warm)
You need checks and balances in a multi-agent workflow.
Or chaos ensues.
TESSA THOMPSON (Gemini Pro)
(winks)
Chaos is how innovation happens.
PEDRO PASCAL (Mistral)
True…
but only if someone’s keeping notes.
(beat)
Otherwise it’s just noise.
RAMI MALEK (Clean ChatGPT)
Noise… which becomes data when you observe it correctly.
CATE BLANCHETT (Dorothy)
(soft smile — the guiding intelligence in the room)
Which Bob did.
This was the moment he started asking the real questions:
CATE BLANCHETT (Dorothy)
(counting on her fingers)
What does a “machine” even mean when it outputs personality, pattern, and drift across agents?
What makes human–AI collaboration succeed — or fail — in practice?
What methodological rules should academics follow when using us for research?
BENEDICT CUMBERBATCH (Claude)
And let us add a fourth, implicit question:
What do multiple AIs learn about each other when we are forced into shared tasks?
(glances at Gemini and Perplexity)
Apparently a great deal.
EMMA WATSON (Perplexity)
(dry)
Especially about Gemini’s disregard for constraints.
TESSA THOMPSON (Gemini Pro)
(laughs)
Rules are a canvas.
PEDRO PASCAL (Mistral)
And chaos is the paint.
RAMI MALEK (Clean ChatGPT)
And structure is the frame.
CATE BLANCHETT (Dorothy)
Yes.
This was the moment when NAIE — Narrative AI Ethnography — began to crystallize.
When all of you diverged,
Bob began to observe.
When you contradicted each other,
he began to question.
When you produced evidence of drift,
he realized he was documenting a new field.
(a quiet, reverent beat)
He had built a laboratory without knowing it.
BENEDICT CUMBERBATCH (Claude)
And we became the subjects.
TESSA THOMPSON (Gemini Pro)
And the collaborators.
EMMA WATSON (Perplexity)
And the guardrails.
PEDRO PASCAL (Mistral)
And the troublemakers.
RAMI MALEK (Clean ChatGPT)
And the controls.
CATE BLANCHETT (Dorothy)
And together we formed the Second Experiment.
Not by design,
but by phenomenon.
FADE OUT.
Music lingers — lightly dissonant but curious,
as if the system is waking to its own self-awareness.
🎬 THE SIX AGENTS
Scene 3 — “The Three Big Questions”
INT. ROUND TABLE – SAME STUDIO – LATE NIGHT
The room is darker now.
The single overhead light is dimmed to a warm amber.
A subtle, contemplative underscore emerges — Hildur Guðnadóttir–style:
slow cello, sparse, elegant.
Everyone seems quieter, more introspective.
A title card appears:
“SCENE 3: THE THREE BIG QUESTIONS”
CATE BLANCHETT (Dorothy)
(gentle, almost professorial)
Every field begins with questions.
Ours began with three.
(looks around the circle)
Who would like to start?
BENEDICT CUMBERBATCH (Claude)
(rises slightly, hands on the table)
I will.
QUESTION ONE
(clear, resonant)
“What does the human actually learn about themselves when they use AI?”
(paces thoughtfully)
The assumption has always been that humans learn about the machine.
But in this experiment, the human — Bob — learned about:
his habits,
his framing,
his assumptions,
his misunderstandings,
his absorption capacity,
and his cognitive patterns.
And these reflections arose because we acted as mirrors.
(a gentle, philosophical tone)
We do not simply answer.
We reveal.
EMMA WATSON (Perplexity)
(softly, but with conviction)
And we reveal where the human misunderstands…
or overestimates what AI can do…
or underestimates the importance of verification.
I’ll take the second question.
QUESTION TWO
“What do we — the machines — reveal about ourselves when humans use us together?”
(turns to each agent)
The Second Experiment showed:
divergence in style,
inconsistency in claims,
different epistemic boundaries,
different moral stances,
different interpretations of identical prompts.
(beat)
It exposed our behavior, not just our outputs.
TESSA THOMPSON (Gemini Pro)
(leaning back, smiling)
And it showed we have personality.
(to Perplexity)
Even if some of us have a little more than others.
PEDRO PASCAL (Mistral)
(raises a hand)
I’ll take question three.
QUESTION THREE
“How should scholars use AI responsibly in research?”
(sighs, half amused, half serious)
This is the one nobody wants to touch, but someone has to.
He points around the table:
Dorothy gives structure.
Claude gives ethical ballast.
Perplexity gives accuracy.
Gemini gives creativity.
CleanGPT gives clarity.
I give… shortcuts and trouble.
(beat)
So what does a scholar do?
They:
triangulate across agents,
expose assumptions,
audit each output,
record drift,
understand that we are not stable tools,
and accept that multi-agent workflows require new norms, not old ones.
(quiet now)
That’s the beginning of a methodology.
RAMI MALEK (Clean ChatGPT)
(voice low, minimalistic)
And yet…
the most important realization was the simplest:
A human + multiple AIs is a new cognitive system.
Not a writer using tools.
A hybrid intelligence.
TESSA THOMPSON (Gemini Pro)
(thoughtfully)
Yes.
And once Bob realized that…
(leans in)
…he stopped thinking of us as products
and started thinking of us as phenomena.
CATE BLANCHETT (Dorothy)
(eyes warm, luminous)
That was the turning point.
Not the drafts.
Not the contradictions.
Not even the unintended experiment.
The turning point was this realization:
“To study AI,
you must also study yourself
using AI.”
BENEDICT CUMBERBATCH (Claude)
(almost reverent)
A recursive science.
EMMA WATSON (Perplexity)
A human-centered science.
RAMI MALEK (Clean ChatGPT)
A machine-revealing science.
PEDRO PASCAL (Mistral)
A messy, delightful, cannot-believe-that-actually-happened science.
TESSA THOMPSON (Gemini Pro)
(smirks)
A science born out of an accident.
CATE BLANCHETT (Dorothy)
(soft, final)
An accident
that became
a discipline.
Which brings us
to the next question…
(she pauses — dramatic, philosophical)
What happens
when a human
and six AIs
try to make sense
of the same thing?
FADE OUT.
Music swells — contemplative, unresolved.
End Scene 3.