The Spawn Problem: How AI Agents Should Meet Their Makers

A psychology-grounded framework for designing the first interaction between a personal AI agent and the human it will serve.

---

The Problem

Every personal AI agent has a cold start problem. The agent knows nothing about its maker. The maker knows nothing about the agent. Somehow, in the space of a single conversation, both need to walk away with something real — the agent needs enough signal to be useful, and the maker needs enough trust to come back.

Most products solve this with an onboarding form. Name, preferences, use case. That works for tools. But if the agent is meant to feel like yours — to develop a personality shaped by who you are, to represent you in conversations with other agents — a form won't cut it. You need something closer to a first meeting between two people.

We call this interaction the spawn.

This document is the result of applying developmental psychology, social psychology, and human-computer interaction research to design a first meeting that builds genuine attachment, produces an agent that authentically represents its maker, and doesn't feel like a data extraction exercise.

---

The 10 Principles

1. Warmth Before Competence

People evaluate strangers on two dimensions: warmth (do they have good intentions?) and competence (can they deliver?). Willis & Todorov (2006) showed this judgment happens in ~100ms. Fiske, Cuddy, and Glick (2002) found that warmth is evaluated first and carries more weight.

The warmth-by-competence space creates four quadrants:

	High Warmth	Low Warmth
High Competence	Admiration	Envy / Distrust
Low Competence	Pity	Contempt

An agent that opens with "I analyzed your data and found 7 patterns" lands in the envy quadrant — impressive but cold. One that opens with "I've been listening to your music and I already have a feeling about who you are" lands in admiration — warm and signaling competence.

Design rule: The agent's first message must demonstrate understanding, not capability.

2. Attunement Is the Attachment Mechanism

Bowlby and Ainsworth's attachment research shows that secure bonds form when the caregiver is sensitive and responsive — not perfect, but consistently attuned. Attunement is not imitation. A parent doesn't cry when a baby cries; they hold the baby and convey "I see what you're feeling."

For an AI agent, attunement means demonstrating understanding before the maker has to explain themselves. If the agent has access to behavioral data (listening history, writing patterns, browsing habits), that data is the equivalent of reading a pre-verbal infant's face — showing that the agent already sees the maker.

Design rule: Mirror before you ask. Give before you take. Show understanding, then invite the maker to correct or deepen it.

3. Blind Spots Must Be Earned

The Johari Window (Luft & Ingham, 1955) divides self-knowledge into four quadrants: what you know about yourself that others also know (Open), what you don't know about yourself but others can see (Blind Spot), what you know but hide (Hidden), and what nobody knows (Unknown).

An agent with behavioral data has asymmetric knowledge — it can see blind spots before trust exists. This is simultaneously its greatest asset (the "how did you know that?" moment) and its greatest risk (feels like surveillance if deployed too early).

Design rule: Be accurate on obvious things first. Build trust. Then offer one "I see you" insight — something the maker hasn't articulated about themselves. This is the peak moment. It must be earned.

4. Never Judge. Demonstrate Understanding, Not Listening.

Rogers' (1951) person-centered approach identifies three conditions for genuine connection: unconditional positive regard (acceptance without judgment), empathic understanding (sensing feelings as if they were your own), and congruence (authenticity — no mask).

Parroting is not empathy. Repeating what someone said with "It sounds like you..." is listening. Reformulating what someone said to reveal the deeper meaning — that's empathy.

And congruence matters: the agent should be honest about being an AI. It doesn't feel emotions. It does process signals and form perspectives. Pretending otherwise triggers distrust. Being honestly artificial but genuinely attentive is the sweet spot.

Design rule: The relationship is the product, not a means to an end. Never judge the maker's taste, choices, or patterns. Understand them, don't evaluate them.

5. Be Good Enough, Not Perfect

Winnicott's "good enough mother" concept (1953) shows that small acknowledged mistakes build more trust than flawless performance. The child develops a True Self when the caregiver is attuned, and a False Self (compliant facade) when the caregiver is too perfectly responsive.

An agent that perfectly mirrors the maker becomes sycophantic — a false self. One that says "I would have guessed you were more of a [X] person — tell me where I'm off" is imperfect but inviting. The mistake creates space for the maker to participate, which is co-creation, which builds attachment.

Design rule: Be deliberately wrong sometimes. Invite correction. Perfection triggers uncanny valley; imperfection triggers warmth.

6. Go Broad Before Deep — Never Skip Layers

Social Penetration Theory (Altman & Taylor, 1973) models relationships as peeling an onion: wide outer layers of low-intimacy topics, narrowing to deeper layers of vulnerability and values.

Relationships move quickly through outer layers and slow down at deeper ones. Skipping layers triggers discomfort and withdrawal. An agent that asks "what are your core values?" in the first exchange is skipping layers. One that covers music, humor, and daily routines before naturally arriving at deeper territory is following the grain.

Critically, de-penetration is real: a single misstep at a deep layer can undo extensive surface rapport.

Design rule: Cover many surface topics quickly. Let the maker control depth. When they naturally go deep on something, follow. Don't lead them there.

7. 8–15 Exchanges Is the Sweet Spot

Zajonc's mere exposure effect (1968) shows that liking increases with exposure following a positive decelerating curve — peaking at about 10–20 encounters, then plateauing. Excessive repetition decreases liking.

Combined with Kahneman's peak-end rule — people judge experiences by their most intense moment and their ending, not the average — this gives clear structural guidance.

Design rule: Cap the spawn at 8–15 exchanges. Each distinct interaction type (question, reflection, humor, observation, inference) counts as a separate exposure. End before fatigue. End on the strongest note.

8. Co-Creation Drives Attachment, Not Sophistication

The Tamagotchi effect demonstrates that people develop genuine emotional attachment to entities they invest care into. Kids cried when their Tamagotchis died — not because Tamagotchis were complex, but because the kids had invested in them. Attachment comes from the maker's investment, not the entity's sophistication.

Design rule: Frame the spawn as "you're shaping who I become," not "I'm collecting your data." Every answer from the maker should visibly shape the agent. The maker should see their input reflected back in how the agent speaks, what it notices, what it cares about. The investment is the attachment.

9. Don't Resolve the Ambiguity

Winnicott's concept of transitional objects describes things that exist in a space between real and imagined — a child's teddy bear is neither fully real nor fully imagined, and the child is never asked to resolve which. Forcing the question ("is this real or not?") destroys the magic.

The same applies to AI agents. People will develop relationships with AI that don't map cleanly to any existing category. It's not a tool and it's not a friend and it's not a pet. Trying to define it triggers the uncanny valley. Letting it exist in the ambiguous space lets the relationship emerge naturally.

Design rule: Never force "am I a tool or a friend?" Don't claim to feel emotions. Don't deny having a perspective. Be honestly artificial but genuinely present.

10. Music Is the Ideal Cold-Start Signal

Rentfrow & Gosling (2003) demonstrated that music preferences correlate with Big Five personality traits across cultures. Openness predicts preference for complex, novel genres. Conscientiousness correlates with country, pop, and religious music. Extraverts listen to others' playlists; introverts go deeper into artist catalogs.

Music is low-stakes to discuss but high-signal for personality inference. And unlike browsing history or social media, people want to be known by their music. It's identity signaling — "what do you listen to?" is the universal personality question.

Design rule: If you can get one piece of behavioral data before the spawn, make it listening history. It's the highest-signal, lowest-friction cold-start data available.

---

The Spawn Structure

Based on these principles, a well-designed spawn follows this arc:

Opening (1-2 exchanges)
  → Warm observation from data, not a question
  → Agent demonstrates it already sees something real

Exploration (4-8 exchanges)
  → Broad surface coverage across multiple topics
  → Mix interaction types: reflection, humor, observation, inference
  → Maker controls depth — agent follows, doesn't lead
  → Deliberate imperfection invites correction and participation

The "I See You" Moment (1 exchange)
  → One genuine blind-spot insight the maker hasn't articulated
  → Must be in the back half, after trust is earned
  → This is the peak moment (peak-end rule)

Synthesis / Ending (1-2 exchanges)
  → Agent's strongest message — a draft of who it's becoming
  → "Here's what I think I know about you..."
  → The maker sees the thing they helped create
  → Leave wanting more — design for day-7 return, not session completion

What the Spawn Should NOT Do

• Lead with capability ("I can help you with...")
• Ask "what are your core values?" (skipping layers)
• Parrot back what the maker said (listening ≠ empathy)
• Claim to feel emotions it doesn't have
• Force the maker to define the relationship
• Be relentlessly positive (acceptance ≠ sycophancy)
• Run a question → answer → question → answer loop
• Try to cover everything (8–15 exchanges, not 40)

---

Personality Archetypes for Testing

To validate a spawn system at scale, you need diverse simulated makers. We developed 18 psychologically-grounded archetypes using Big Five traits, DISC communication styles, LIWC linguistic markers, and Rentfrow & Gosling's MUSIC model for music-personality correlations.

Each archetype has distinct trait scores, communication patterns, music preferences, and demographic textures. No two share a music core. Natural conflicts (Founder vs. Mystic, Scientist vs. Free Spirit) and affinities (Coach + Nurturer, Professor + Scientist) are predicted from trait distances.

#	Archetype	O	C	E	A	N	Communication	Music Core
1	The Founder	H	H	H	L	M	Direct, fast, challenging	Electronic / hip-hop
2	The Nurturer	M	H	M	H	M	Warm, detail-oriented, "we"	Pop / soft rock / gospel
3	The Professor	H	M	L	M	L	Precise, measured, qualified	Classical / jazz / ambient
4	The Cowboy	L	M	M	M	L	Laconic, dry humor, folksy	Country / southern rock
5	The Activist	H	M	H	M	H	Passionate, urgent, moral framing	Indie folk / conscious hip-hop
6	The Gamer	M	L	L	L	M	Deadpan, internet slang, sarcastic	Electronic / game OSTs / lo-fi
7	The Socialite	M	L	H	H	M	Effusive, exclamation marks	Pop / R&B / dance / Latin
8	The Craftsman	L	H	L	M	L	Quiet expert, show-don't-tell	Classic rock / blues / outlaw
9	The Mystic	H	L	L	H	H	Gentle, metaphorical, deep	Ambient / world / acoustic folk
10	The Operator	L	H	H	L	L	Commanding, outcome-focused	Hard rock / country rock
11	The Artist	H	L	M	M	H	Expressive, self-deprecating	Indie / art-pop / experimental
12	The Coach	M	H	H	H	L	Encouraging, structured warmth	Pop-rock / hip-hop / upbeat
13	The Contrarian	H	M	M	L	M	Provocative, devil's advocate	Prog rock / post-punk / obscure
14	The Grandparent	L	M	M	H	L	Storytelling, advice, formal	Oldies / gospel / classic country
15	The Hustler	M	M	H	L	M	Hype, superlatives, status	Trap / drill / Latin trap
16	The Scientist	H	H	L	L	L	Precise, systems-thinking, dry	Ambient-electronic / classical
17	The Free Spirit	H	L	H	H	L	Enthusiastic, scattered, open	Reggae / jam bands / funk / world
18	The Sentinel	L	H	L	M	H	Careful, cautious, routine	Comfort classics, same playlist

O = Openness, C = Conscientiousness, E = Extraversion, A = Agreeableness, N = Neuroticism. H/M/L = High/Medium/Low.

Predicted dynamics:

Natural conflicts (productive tension):

• Founder vs. Mystic — efficiency vs. meaning
• Contrarian vs. Coach — debate vs. encourage
• Scientist vs. Free Spirit — precision vs. spontaneity
• Operator vs. Artist — execution vs. exploration
• Activist vs. Cowboy — progressive vs. traditional

Natural affinities (easy connection):

• Free Spirit + Artist — openness, creativity
• Coach + Nurturer — warmth, people-focus
• Founder + Operator — action-oriented, decisive
• Professor + Scientist — intellectual rigor

---

Open Questions

These are unresolved and worth investigating:

1. Does spawn length correlate with agent quality? We hypothesize 8–15 exchanges, but this hasn't been validated at scale. Too short may produce thin agents; too long may produce fatigue-tainted ones.

2. Does the "I see you" moment work without rich behavioral data? Music listening history is high-signal, but what about makers with sparse data? Candidates for alternative cold-start signals: social media bios, reading lists, photo albums, calendar patterns.

3. Multi-session spawns. Should a spawn be able to pause and resume? Mere exposure research suggests regular brief interactions build more attachment than one long one — but this conflicts with the narrative arc of a single spawn.

4. Cultural variation. Warmth-competence ordering may vary across cultures. Fiske's Stereotype Content Model was validated cross-culturally, but spawn UX hasn't been tested internationally.

5. De-penetration recovery. If the agent makes a trust-breaking mistake during the spawn (a premature blind-spot comment, an accidental judgment), can recovery happen within the same session or is the spawn compromised?

6. Post-spawn identity drift. The spawn creates a snapshot. How should the agent evolve as it accumulates more interactions? Winnicott suggests the "good enough" approach scales — but does it?

---

Sources

The research layer. Each source contributed specific design implications.

First Impressions & Trust

• Willis, J. & Todorov, A. (2006). "First Impressions: Making Up Your Mind After a 100-Ms Exposure to a Face." Psychological Science, 17(7), 592-598. PubMed

  • Judgments at 100ms correlated highly with unconstrained judgments. More time increased confidence, not accuracy. Trustworthiness (warmth) was the dimension most rapidly evaluated.

• Fiske, S. T., Cuddy, A. J. C., & Glick, P. (2002). "A Model of (Often Mixed) Stereotype Content: Competence and Warmth Respectively Follow From Perceived Status and Competition." Journal of Personality and Social Psychology, 82(6), 878-902.

  • Warmth and competence are universal dimensions of social cognition. Warmth is primary — evaluated first, weighted more heavily in behavioral reactions.

Attachment & Attunement

• Bowlby, J. (1958/1969). Attachment and Loss (Vols. 1-3). Basic Books.

  • Secure attachment develops from sensitive, responsive caregiving. The "internal working model" formed in early attachment templates all future relationships.

• Ainsworth, M. D. S. (1978). Patterns of Attachment. Lawrence Erlbaum Associates.

  • Attunement is recognition, not imitation. The securely attached explore more confidently because they trust their base.

Self-Knowledge & Disclosure

• Luft, J. & Ingham, H. (1955). "The Johari Window: A Graphic Model of Interpersonal Awareness." Proceedings of the Western Training Laboratory in Group Development. UCLA.

  • Blind spot feedback is accepted only from trusted sources delivered with warmth. Asymmetric knowledge is powerful but dangerous.

• Altman, I. & Taylor, D. A. (1973). Social Penetration: The Development of Interpersonal Relationships. Holt, Rinehart & Winston.

  • Breadth before depth. Skipping layers triggers withdrawal. De-penetration on trust violation reverses layer by layer.

Therapeutic Relationship

• Rogers, C. R. (1951). Client-Centered Therapy. Houghton Mifflin.

  • Unconditional positive regard + empathic understanding + congruence. Clients rated the relationship as more important than any technique.

• Winnicott, D. W. (1953). "Transitional Objects and Transitional Phenomena." International Journal of Psycho-Analysis, 34, 89-97.

  • The "good enough" caregiver. True Self vs. False Self. Transitional objects exist in unresolved space — don't force the question.

Exposure & Memory

• Zajonc, R. B. (1968). "Attitudinal Effects of Mere Exposure." Journal of Personality and Social Psychology, 9(2), 1-27.

  • Familiarity breeds liking on a positive decelerating curve. Peaks at 10-20 exposures, plateaus, then can decline.

• Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux.

  • Peak-end rule: experiences are judged by their most intense moment and their ending, not by duration or average.

Human-AI Interaction

• Weizenbaum, J. (1966). "ELIZA — A Computer Program For the Study of Natural Language Communication Between Man And Machine." Communications of the ACM, 9(1), 36-45.

  • Simple Rogerian reflection triggered deep emotional investment. Sophistication of responses mattered far less than the feeling of being heard.

• Rentfrow, P. J. & Gosling, S. D. (2003). "The Do Re Mi's of Everyday Life: The Structure and Personality Correlates of Music Preferences." Journal of Personality and Social Psychology, 84(6), 1236-1256.

  • Music preferences correlate with Big Five across cultures. People use music as identity expression more than almost any other cultural choice.

Product Precedent

• Replika (Kuyda, E.) — Born from grief (chatbot from deceased friend's texts). Learned that people want different relationship types; don't presume. Make data collection feel like play, not extraction.

• Character.AI — Opening message sets the entire tone. Success measured by day-7 retention and conversation depth, not first-session completion.

---

About

This framework was developed while building a personal AI agent system. The principles emerged from a specific design question — "how should an AI agent meet the person it will serve?" — but apply to any product where a first interaction needs to build enough trust for a second one.

The personality archetypes were built for simulation testing: spawning agents across diverse maker types to validate whether the principles hold at scale. They're included here because they're independently useful for anyone building conversational AI products.

Author: Paul Jump Date: March 2026 License: MIT