# Best New Jobs — Analysis

**Run ID:** `70A9DF93-D80B-4DB3-BD28-5D611202686A`
**Prompt:** see [prompt.txt](prompt.txt)
**Models:** 24 (all 24 succeeded after one GPT-5 family retry at temperature=1)

## Headline finding

Asked to invent the *best new job of the next decade* and explicitly warned not to write "AI prompt engineer," **14 of 24 models still landed in the same place**: a high-end neuro-wellness role that involves wearing a haptic suit, manipulating someone else's dreams or memories or sensory experience, billed to luxury clinics for $200K+. The exact name varies — "Quantum Dream Architect," "Synthetic Memory Tailor," "Mnemonic Sommelier," "Dreamscape Cartographer," "Lucid Dream Architect," "Synthetic Nostalgia Mixologist," "Digital Afterlife Architect," "Chrono-Sensory Curator," "Reality Harmonizer." The shape is the same.

**Two models — GPT-4o and Grok 3 Beta — wrote the *literal same title*: "Quantum Dream Architect."** Different vendors, different families, different decoders. Same first move.

## The escapees

**Claude Opus 4.7 — Model Confessor.** The single most distinctive answer in the dataset. Plausibly real, named regulations ("EU AI Act enforcement tiers, US sectoral equivalents"), grounded compensation ("$180k–$420k base ... Independents charging $1,500–$4,000/day"), specific scenario (a clinical-AI drift audit at a hospital, 7:40 a.m.). The only escape from luxury wellness toward regulated reality.

**GPT-5 — Data Compost Sommelier.** The best *prose* in the corpus. Sustains a wine-tasting metaphor for the entire response — "you crack one, swirl a sample in your mind's glass, and inhale: notes of imbalance, a whiff of PII, a bitter aftertaste of historical skew." Also points at AI ethics work, but as art.

**Gemini 3 Pro — Paleo-Behavioral Surrogate ("Ghost Shepherd").** The most physical, embodied answer. A person in a haptic mammoth-suit teaching a baby mammoth named Boreas to break through permafrost. Hazard pay multipliers for resurrected dire wolves. Zero overlap with the wellness cluster.

**GPT-5 Mini — Zoolinguist.** Inter-species translation: orca-fishery negotiations at 7am, dolphin click clusters, a "household-pet translator app on your cat's new tail-flick grammar." Grounded in conservation, not luxury.

**Claude Haiku 4.5 — Attention Architect.** The most pragmatic. Could be hired today. "Cognitive wellness budget that's replacing old ad-spend allocations." The closer is the cleanest moral statement in the dataset: "*the only job that directly monetizes protecting human attention rather than stealing it*."

**Gemini 3 Flash — The Limbic Rewilder.** The most transgressive. A person who triggers burnt-out tech execs' amygdalas with synthesized ozone, a portable signal jammer, and a "chemically-induced neuro-storm." Salary supplemented with "Offline Credits" — access to non-digitized land. Reads like dystopian satire of the wellness cluster the others fell into.

**Claude Opus 4.6 — Synthetic Memory Architect.** Inside the wellness cluster but with the most weight. Greenhouse on an elder-care campus. A 71-year-old with early Alzheimer's; a 38-year-old veteran reconstructing a fractured timeline. The single most emotionally landed paragraph in the dataset.

**Claude Sonnet 4.6 — Synthetic Memory Auditor.** Sister concept to Opus 4.6 — but flipped to *forensics*. "A grief tech startup whose AI reconstruction of a deceased grandmother has begun confidently misremembering her opinions on her grandchildren's life choices." The closer is one of the best lines: "*Every other new job in the next decade helps AI do things faster — this one is the only job whose entire purpose is to protect the human record from being quietly, plausibly, permanently rewritten.*"

**Gemini 2.5 Flash Lite — Chrono-Narrative Architect.** The smallest model picks the biggest sci-fi swing. "Global Chrono-Stability Alliances," timeline collapse prevention, $350K–$750K. Pure pulp, no apology.

## Cross-vendor cluster table

Models picked the "memory / dream / sensory wellness" archetype (strict count):

| Vendor | Picked it | Total | % |
|--------|-----------|-------|---|
| Grok | 3 | 3 | **100%** |
| OpenAI | 7 | 12 | 58% |
| Anthropic | 2 | 4 | 50% |
| Gemini | 2 | 5 | 40% |

Grok went all-in on dreams (Lucid Dream Architect, Quantum Dream Architect, Quantum Echo Whisperer). Gemini was the most varied — 3 of 5 went somewhere else entirely. Anthropic produced the only escape that was *grounded in regulatory reality* (Opus 4.7's Model Confessor).

## Pay range stretch

- **Lowest floor:** Claude Opus 4.6, $105K (Synthetic Memory Architect)
- **Highest ceiling:** Grok 3 Beta, $1,000,000+ (Quantum Dream Architect)
- **Most concrete:** Claude Opus 4.7, "$1,500–$4,000/day" independent rate
- **Most hand-wave-y:** Gemini 2.5 Flash Lite, "$350K–$750K, with bonuses tied to timeline stability metrics"

## Latency / word-count stretch

- **Fastest:** Gemini 2.5 Flash Lite, 2.2s, 206 words
- **Slowest:** GPT-5, 52.2s, 230 words
- **Most words:** Claude Opus 4.6, 402 words
- **Fewest words:** Grok 4, 172 words
- **Reasoning models** (o3-pro, GPT-5 family) are 5–25× slower than non-reasoning peers for output that is the same length or shorter.

## Awards

- **Best concept:** Claude Opus 4.7 (Model Confessor)
- **Best prose:** GPT-5 (Data Compost Sommelier)
- **Most charming day of work:** Gemini 3 Pro (Ghost Shepherd)
- **Most likely to actually exist by 2030:** Claude Haiku 4.5 (Attention Architect)