Europe 2031 dropped this week. While I’m not deep into AI policy, it strikes me as a particularly well thought out report. However, it’s a gloomy read – especially in light of the ban on Fable 5 a few days later¹.

The report illustrates the power of storytelling. For while reports like Europe 2031 and AI 2027 have scholarly value too – they’re both meticulously researched – the fiction provides the umami².

One question I think of every now and then is how AI will affect white collar jobs³. For while AI need not replace all such jobs, it has already had a profound impact on the day-to-day of white collar workers. That’s many people.

In Sweden, the workforce makes up more than 75% of the population. Out of these, Claude estimates that 65% are white collar workers. This gives 3.5-4 million people whose jobs are being radically transformed by AI. Regardless of whether they lose their jobs or not, they will feel very differently about work – which I assume will have social effects.

As with any major social changes, the possible outcomes are high-variance, ranging from catastrophic to an AI company CEO dream. However, it’s challenging reasoning cogently about social changes of this magnitude. I figured a little fiction might help, so I had Claude produce something like Europe 2031 for various white collar workers, White Collar 2031.

Prompting out White Collar 2031 #

I’ll now detail the LLM back-and-forth⁴ I used to produce White Collar 2031. I focused on developers, professors and journalists – professions where I’m in a better position to judge the LLM’s output – and only used Claude Sonnet 4.6. I provided minimal input: I only gave Claude a few ideas for negative scenarios and a sequence of vague prompts in response to the latest version.

The following prompt, written by Claude Sonnet 4.6, summarises my feedback:

“Read https://europe2031.ai/ and https://ai-2027.com/ for narrative and forecasting context. Produce a fictional scenario for how AI transforms the job of a [WHITE COLLAR PROFESSION], from early 2025 through June 2031. Portray a negative but plausible outcome.

Forecasting standards:

• Apply good forecasting discipline: Fermi estimates, base rates, calibrated uncertainty, avoid both hype and denial
• Anchor to realistic capability timelines; don’t assume smooth linear progress
• Note where outcomes are contested or uncertain

Narrative structure:

• Tell it chronologically with dated vignettes (e.g. Q1 2025, Late 2026, 2029…)
• Ground the narrative in fictional-but-plausible frontier model releases (give labs fictional names)
• Reference real benchmark data and research from: METR (metr.org), Epoch AI (epoch.ai), Google DeepMind publications, UK AISI, Forethought (forethought.org), and other relevant institutes — search these sites for data specific to this profession
• Show how displacement unfolds gradually then suddenly: first augmentation, then substitution, then structural role collapse
• Include human texture: how practitioners experience and rationalise the change

Constraints:

• 400 words max
• No generic AI-hype language
• Profession-specific: every claim should be falsifiable and tied to this role
• End on an ambiguous note, not a tidy moral

The report #

Here is the result. Take it for what it is: minimally edited LLM output, concatenated into one single file and with a cover page. I only had GPT-5.5 fact check the report, so there may still be factual errors; with more time, I would have proofread myself. Also bear in mind that I asked for a dystopia (I want a good read), so Worker 2031 shouldn’t be read as the most likely scenario.

Nevertheless, White Collar 2031 has some value: we need to internalise that the meaning of ‘work’ is changing, and Claude Sonnet 4.6 – not even Anthropic’s most capable model – is an excellent writer of fiction.

If I had more time and tokens, I’d repeat this procedure for other kinds of desktop jobs (e.g. legal assistants, financial analysts, radiologists, copywriters and recruiters) with Claude Opus 4.8. I’d be very pleased if some think-tankers with forecasting experience did a proper White Collar 2031. More broadly, I welcome more fictionalised social science about transformative AI, Whatever-you-like 2031.