💡 Full disclosure: I told Claude Sonnet 4.6 to build this. My prompt: "Build a web app showing how AI will take white-collar jobs over time — visualize it, publish it online, write articles about it." The simulation, code, deployment, and these articles were all produced by Claude in one session.
💡 Full disclosure: I told Claude Sonnet 4.6 to build this. My prompt: "Build a web app showing how AI will take white-collar jobs over time — visualize it, publish it online, write articles about it." The simulation, code, deployment, and these articles were all produced by Claude in one session.
Explore the full interactive simulation at sim-wine.vercel.app
When people talk about AI taking jobs, the conversation is usually binary: either AI will take all jobs, or it won't. The reality is far more structured — specific roles get hit at specific times, and the timing matters enormously.
I built a simulation using sigmoid (S-curve) displacement models to map out when each major white-collar job category gets disrupted. Here's the math and the data.
Why Displacement Follows an S-Curve
Technology adoption doesn't happen linearly. It follows a sigmoid pattern:
- Slow start: early adopters, high cost, limited AI capability
- Rapid acceleration: widespread adoption, falling cost, dramatically improving models
- Plateau: structural limits, regulatory friction, irreducible human preference
For AI job displacement, the formula is:
displacement(year) = maxDisplacement × sigmoid(k × (year − inflectionYear))
Where sigmoid(x) = 1 / (1 + e^(−x)) and k controls how steep the curve is.
Three Parameters That Determine Your Job's Fate
Maximum Displacement — the ceiling, not the destination. Physical, legal, or preference constraints cap automation. Even data entry — the most automatable role — caps at 97%, not 100%, because edge cases always require humans.
Inflection Year — when disruption hits fastest. Before this point: gradual nibbling at tasks. After: rapid displacement of full roles. Early-wave inflection points cluster around 2027–2030. The most resistant roles don't hit their inflection until 2039–2041.
Transition Rate — what fraction of displaced workers successfully pivot to AI-augmented or new roles. Software developers have the highest transition rate (58%) because their underlying problem-solving skills transfer. Data entry clerks have the lowest (10%) — the skills are narrow and the AI replacement is complete.
The Full Dataset: 20 Jobs, 24 Years
| Job Category | Workers (2026) | Max Displacement | Inflection Year | Transition Rate |
|---|---|---|---|---|
| Data Entry Clerks | 320K | 97% | 2027 | 10% |
| Tax Preparers | 75K | 92% | 2030 | 13% |
| Bookkeepers | 1.2M | 86% | 2030 | 25% |
| Admin Assistants | 3.4M | 83% | 2030 | 22% |
| Insurance Underwriters | 100K | 80% | 2032 | 20% |
| Customer Service Reps | 2.9M | 79% | 2030 | 18% |
| Copywriters & Content | 280K | 74% | 2029 | 30% |
| Research Analysts | 450K | 72% | 2033 | 28% |
| Paralegals | 350K | 72% | 2032 | 25% |
| Loan Officers | 320K | 69% | 2033 | 22% |
| Marketing Analysts | 850K | 67% | 2034 | 32% |
| Financial Analysts | 500K | 64% | 2034 | 35% |
| Journalists & Reporters | 60K | 62% | 2032 | 35% |
| HR Specialists | 650K | 59% | 2034 | 30% |
| CPAs & Auditors | 1.3M | 58% | 2035 | 38% |
| Business Analysts | 700K | 56% | 2036 | 40% |
| Software Developers | 4.6M | 53% | 2037 | 58% |
| Project Managers | 700K | 43% | 2039 | 46% |
| Lawyers | 900K | 39% | 2039 | 42% |
| Management Consultants | 600K | 36% | 2041 | 52% |
Sources: BLS OES 2024 (worker counts), McKinsey Global Institute 2023, Goldman Sachs Research 2023, Stanford HAI Index 2024, Oxford Future of Employment
What Makes a Role Resistant to Automation?
Looking at what separates the high-resistant roles (lawyers, consultants, project managers) from the high-displacement ones:
1. Physical presence + trust relationships
Roles where clients pay for human accountability — where being wrong carries professional liability — resist automation longest. Lawyers and doctors aren't primarily knowledge workers; they're trusted agents with personal accountability.
2. Genuine ambiguity over structured decisions
Tasks where the "right answer" is contested, context-dependent, or involves value judgments are harder to automate. Routine contract review? Automated. Whether to settle or litigate? That needs judgment.
3. Regulatory friction
High-stakes decisions (medical diagnoses, legal advice, financial recommendations) require human sign-off by law in most jurisdictions. Regulation creates a floor below which automation can't go.
4. Creative synthesis vs. pattern matching
AI excels at retrieving and recombining patterns. Genuine creative synthesis — novel solutions to poorly-defined problems — is harder. Management consultants framing a new market entry strategy, for example, are doing something different from analysts running models.
5. Emotional intelligence as the last frontier
The highest-displacement roles require little emotional intelligence. The lowest-displacement roles are saturated with it: courtroom advocacy, client relationship management, team leadership. AI is improving at emotion recognition, but it's the last frontier.
The Transition Gap: The Real Crisis
The most alarming finding isn't the peak displacement numbers — it's the mismatch between who gets displaced and who gets the new jobs.
Consider administrative assistants: 3.4M workers, 83% peak displacement, only 22% transition rate. That's ~2.1M people with skills that won't command market value in an AI world.
The 6M net new AI-era jobs that emerge by 2050 require:
- Technical comfort with AI tools (prompt engineering, workflow automation)
- Data literacy (interpreting AI outputs, spotting errors)
- Problem-framing ability (not just answering questions, but asking the right ones)
- Cross-domain reasoning (connecting AI capabilities to business problems)
These are precisely the capabilities that administrative work, data entry, and bookkeeping don't develop. The "new jobs will appear" argument is true in aggregate — but the distribution of who gets those new jobs is deeply unequal.
What to Do With This Information
If your role is in Wave 1 (2026–2031):
The timeline is short. Start now. The skills to develop: AI tool proficiency, domain expertise that AI can't easily replicate, management of AI systems in your field.
If your role is in Wave 2 (2031–2035):
You have a window, but it's narrower than it looks. The S-curve means the displacement accelerates sharply at the inflection point. Reskilling takes 2–3 years minimum. Starting in 2033 for a 2034 inflection is cutting it close.
If your role is in Wave 3 (2035–2042):
"Resistant to automation" doesn't mean "immune." The roles that survive are the ones that use AI to multiply their output — not ignore it. A lawyer using AI research tools can handle 3× the caseload. The lawyers who resist AI adoption will be underpriced by those who don't.
See the full interactive simulation with real-time animations at sim-wine.vercel.app
Top comments (0)