Jobs Threatened by AI: Who Loses, Who Adapts, and the Cascade Effects

In the last 18 months, over 500,000 tech employees have been laid off globally. Microsoft, Google, Amazon, Meta, SAP, Duolingo, Chegg — each wave justified by the same formula: “restructuring and investment in AI”. But what does this mean in concrete terms? Who are the people behind the numbers, what happens to them after being laid off, and — more importantly — what happens when the effects cascade through the economy?

This analysis compares the situation in Romania and the United States, identifies the most vulnerable professional categories, examines real data from 2024-2026, and proposes concrete transitions — not slogans.

1. The 2024-2026 Layoff Wave: What Actually Happened

USA: Starting in 2024, Big Tech companies did not merely reduce headcount — they permanently restructured entire teams. Google laid off 12,000 in January 2024, then another 1,000+ from Ads and Cloud in May 2025. Microsoft eliminated 10,000 and rehired in “AI-native” roles. Amazon restructured Alexa, HR, and finance teams — 27,000 combined in 2024. Meta went through three successive rounds. Chegg (education) lost 80% of its stock value after ChatGPT replaced its service. Duolingo dropped 10% of its contractors, replacing them with GPT-4.

Romania: The impact is harder to measure — we don't have a local layoffs.fyi. But the signals are clear: UiPath laid off 10% of its global workforce (~700 employees, including from Bucharest), outsourcing companies (Endava, Ness, Luxoft) have reduced teams on automated projects, and BPO centers in Cluj, Timișoara, and Iași have begun replacing customer support agents with AI conversational solutions. We estimate 10,000-15,000 jobs directly lost in Romania in 2024-2026 in tech + BPO, with another ~40,000 indirectly affected.

2. The Most Threatened Professional Categories

Based on real data from layoffs, OECD and McKinsey reports, and our internal model, the vulnerability hierarchy looks as follows:

Extreme risk (>70% automation in 5 years)

• Data entry and document processing — GPT-4V + IDP (Intelligent Document Processing) already replaces 80%+ of the work. BPO companies in Romania (which processed invoices, contracts, and forms for Western clients) are the hardest hit.
• Customer support tier 1-2 — AI chatbots (Intercom Fin, Zendesk AI, custom solutions) already resolve 60-80% of tickets without human intervention. In Romania: call centers in Cluj, Iași, and Brașov are losing volume massively.
• Translation and localization — DeepL, GPT-4, Claude translate at 90%+ accuracy. Remaining human translators only do revision. The market has contracted by ~40% in 2 years.
• Routine accounting and bookkeeping — platforms like Bench, Pilot, and Ramp automate reconciliations, classifications, and reporting. In Romania: small accounting firms are losing clients to SaaS + AI.

High risk (50-70% automation in 5 years)

• Junior Software Development — Cursor, GitHub Copilot, Claude Code allow a senior to do the work of 3-5 juniors. Companies are hiring fewer entry-level and more “AI-augmented” seniors. In Romania, where ~60% of the IT market is execution-based outsourcing, the impact is enormous.
• Manual QA Testing — AI generates automated tests (Testim, Mabl, Copilot for tests). Manual QA roles disappear; only QA automation engineers are still needed.
• Marketing copywriting and content — Jasper, Copy.ai, ChatGPT reduce content teams from 10 people to 2 people + AI. Marketing agencies in Romania began restructuring in 2025.
• Junior financial analysts — Bloomberg Terminal + GPT, AlphaSense, and other AI tools reduce the need for analysts who compile data and produce standard reports. Goldman Sachs reduced its analyst cohorts by 30%.
• Paralegals and legal assistants — Harvey AI, CoCounsel process contracts, due diligence, and legal research at a fraction of the cost. In Romania: large law firms in Bucharest are already testing these tools.

Moderate risk (30-50% transformation)

• Routine graphic design and UI/UX — Midjourney, DALL-E, Figma AI reduce production time by 60-70%. Designers become “creative directors” of AI.
• News journalism and reporting — Associated Press has been using AI for years for financial reports. Now it is expanding to sports, weather, and local news. Investigative journalism remains human.
• HR screening and initial recruiting — AI filters CVs, schedules interviews, and does pre-screening. Remaining recruiters focus on human evaluation and culture.

Low risk (under 30%)

• Skilled physical trades — electricians, plumbers, mechanics, welders. Robots that can navigate real homes and solve unique problems do not yet exist and will not exist in 5 years.
• Healthcare — doctors, nurses, surgeons — AI becomes a diagnostic tool, but clinical decision-making and intervention remain human. An acute shortage of medical personnel makes layoffs unlikely.
• Education (teachers) — AI becomes a teaching assistant, but the teacher-student relationship remains critical, especially in K-12.
• Personal care — babysitting, elderly care, physical therapy. Essential human contact, impossible to automate.

3. What Happens to the Laid-Off Population? Real Data

We analyzed available data from the Bureau of Labor Statistics (US) and INS/ANOFM (Romania) to understand the post-layoff trajectory:

In the USA:

• 35% are re-employed within 3 months — but often at salaries 15-25% lower, in roles beneath their previous level.
• 25% undergo retraining — bootcamps, courses, certifications. Success rate: ~40% find a relevant job within 12 months.
• 20% accept gig work / freelancing — Uber, DoorDash, Upwork. Unstable income, no benefits (health insurance, pension).
• 15% temporarily exit the labor market — return to studies, move in with parents, live off savings.
• 5% never return — early retirement, emigration, the informal economy.

In Romania:

• The re-employment rate is higher (50% within 3 months) — but the market is smaller, so competition is fiercer and salaries compress more rapidly.
• Retraining is more difficult — the bootcamp offering is limited, quality is uneven, and Romanian employers are more skeptical of retrained candidates.
• Accelerated brain drain — IT professionals laid off in Romania move to Germany, the Netherlands, and the UK. We estimate ~3,000-5,000 net departures in 2025 from the tech sector.
• The safety net is weaker — unemployment benefits are 75% of the average salary over the last 12 months, but capped at 12 months with a ceiling of ~2,800 RON (~560 EUR). Insufficient for a former developer paying rent in Cluj.

4. The Cascade Effect: Why One Lost Job Is Not Just One Lost Job

AI layoffs have a multiplier effect that superficial analyses ignore. A developer laid off in Cluj is not just a -1 in a statistic. It is a chain of consequences:

Level 1: Direct impact

• The laid-off person loses income, reduces spending, cancels subscriptions, postpones purchases.
• The family is affected — stress, uncertainty, migration decisions.

Level 2: Local economic impact

• Restaurants, cafés, and gyms in tech zones lose customers. In San Francisco, restaurant closures in SoMa increased by 40% in 2024-2025.
• The real estate market reacts — rents in Cluj fell by 8-12% in 2025 following IT sector reductions. Landlords lose income.
• Local tax revenues decline — less income tax means less money for infrastructure, schools, and public services.

Level 3: Systemic impact

• Wage deflation — when 10,000 developers simultaneously look for jobs, average salaries fall. Those who remain employed are also indirectly affected — bargaining power decreases.
• Investment declines — startups find it harder to secure funding when the labor market is unstable. VCs become even more cautious in Eastern Europe.
• Consumer confidence falls — even those who keep their jobs spend less (“precautionary savings”). GDP feels the impact.
• Political pressure rises — anti-tech and anti-AI populism gains traction. In Romania, the rhetoric of “foreigners are taking our jobs” transforms into “AI is taking our jobs.” Punitive regulations become likely.

Level 4: The negative feedback loop

• Companies lay off more people → the local economy contracts → companies have fewer customers → they lay off again.
• This cycle is precisely what Psychohistory models as a “social contraction spiral.” History shows it only stops through active intervention (mass retraining, public works projects, fiscal incentives).

5. Romania vs. USA Comparison: Who Is More Vulnerable?

Romania is structurally more vulnerable for three reasons:

1. Dependence on execution, not creation — Romania built an IT industry based on “we do what Western clients tell us, cheaper.” Exactly the type of work AI automates best. The USA loses jobs too, but it creates the jobs of the future (AI researchers, prompt engineers, AI product managers).
2. Lack of a product ecosystem — the USA has Google, Microsoft, OpenAI, Anthropic — the companies building AI. Romania has no globally relevant AI product. UiPath was the closest, but relocated its HQ to New York.
3. Accelerated brain drain — When Romanian IT contracts, talent leaves. But unlike other crises, this one is not cyclical — the jobs do not return when “the economy recovers.” AI is permanent.

The USA has significant advantages in managing the transition:

• A more flexible labor market — re-employment is faster, even if at lower salaries.
• A much more developed retraining ecosystem (Coursera, Udacity, bootcamps).
• New AI jobs are created within the same economy — emigration is not required.
• Capital available for post-layoff entrepreneurship (accelerators, VC).

6. How People Can Make the Transition: A Concrete Guide

Not “learn to code” — that was advice from 2015. In 2026, the strategy is different:

If you are in tech (developer, QA, data analyst):

1. Become “AI-native” — learn to work WITH AI, not against it. A developer using Claude Code / Cursor is 3-5x more productive. Companies will no longer hire developers who refuse to use AI.
2. Move up the value stack — from execution to architecture, from coding to system design, from implementation to technical strategy. AI replaces labor, not judgment.
3. Specialize in areas where AI cannot — cybersecurity, critical infrastructure, embedded systems, technical compliance.

If you are in BPO / customer support / data entry:

1. Transition to AI management — companies need people to train, monitor, and correct AI. “AI trainer,” “conversation designer,” and “quality analyst for AI” are growing roles.
2. Retrain into adjacent roles — from customer support to customer success (relational, strategic); from data entry to data governance or compliance.
3. Consider skilled trades — this is not condescending advice. A qualified electrician in Romania earns 6,000-10,000 RON net and has guaranteed demand for 20+ years. A call center operator faces an 80% automation risk within 3 years.

If you are a student or early in your career:

1. Do not learn to do what AI already does — learning pure coding (without systems understanding) is like learning to use an abacus when calculators exist. Learn to think in systems, define problems, and evaluate AI outputs.
2. Invest in “humanity-proof” skills — negotiation, leadership, emotional intelligence, complex communication. These are the last things AI will be able to do.
3. Build a portfolio of AI-augmented projects — demonstrate that you can use AI as a multiplier, not just as a replacement.

7. What Governments Should Do (But Probably Won't Do in Time)

Romania:

• National AI retraining program — “Digital Romania 2.0” — 50,000 spots/year in publicly funded retraining programs, not generic online courses but practical 6-12 month programs with guaranteed employment.
• Fiscal incentives for companies that retrain — instead of layoffs → internal retraining with state subsidy. Costs less than unemployment + brain drain.
• Equitable taxation of automation — not a “robot tax” (too blunt), but an “AI displacement levy” proportional to the number of roles eliminated and the profit generated by automation. Revenue directed to the retraining fund.
• Investment in vocational education — vocational schools in Romania are underfunded and stigmatized. In Germany, the dual system (school + in-company apprenticeship) produces the most stable workforce in the EU. Romania needs the same model.

USA:

• Expanding the safety net — US unemployment lasts a maximum of 26 weeks, without health insurance. In the AI era, where retraining takes 12-24 months, this is completely inadequate.
• Portability of benefits — employer-tied health insurance is an anchor that impedes mobility. In the gig + AI era, benefits must be individual and portable.
• Universal Basic Income pilot — not as a permanent solution, but as a transition bridge. Sam Altman (OpenAI) funded pilot studies showing mixed but promising results.

8. Psychohistory Predictions: What Comes Next (2026-2030)

Our Monte Carlo model (10,000 iterations, 12 variables, 5-year horizon) suggests:

The Foundation Path scenario (probability ~30%): Romania and the USA implement aggressive retraining programs. AI displacement tax funds retraining. Brain drain stops. New industries emerge (AI services, green tech, biotech). Net job loss is 5-8%, compensated by productivity gains and new roles.

The Empire Path scenario (probability ~70%): Governments react slowly. Retraining is left to the individual. Brain drain accelerates in Romania. The USA absorbs the shock better but inequality increases massively. Net job loss is 12-18%. Anti-tech populism wins elections in 2-3 European countries. Punitive regulations slow AI adoption in the EU but do not stop layoffs (companies move operations to the US/Asia).

“Do not ask whether AI will take your job. Ask whether someone who uses AI will take your job. The answer is almost certainly yes — and the question becomes: will that someone be you, or someone else?” — olivLaw Psychohistory

9. The Second Foundation: 5 Seeds That Propagate on Their Own

“The actions of the Second Foundation were never blunt. They were small, precise, and placed at the points of maximum leverage — so that the system would do the rest of the work itself.”

In the logic of Psychohistory, an effective intervention is not a massive one — it is one that reaches critical mass, after which it propagates through the network without additional effort. Each action below is designed as a seed: it requires a small initial impulse, but once it takes hold, it generates a self-sustaining network effect.

The principle: you do not convince 1,000,000 people. You convince 500 people who each convince 20.

Seed 1: “The Invisible Mentor” — Peer-to-peer retraining network

The propagation mechanism: A senior developer in Bucharest mentors a laid-off QA from Brașov to transition to AI testing. When the QA finds a job, they post on LinkedIn: “I made it in 4 months, thank you @mentor.” The post is seen by 5,000 people. 30 of them also ask for a mentor. 10 of the previous graduates agree to become mentors. The network doubles every 2-3 months with no additional investment. After 12 months, you have 8,000-10,000 active pairs.

Why it works: Advice from a peer (“I was like you 6 months ago”) is 4-7x more credible than a formal course. Cost: ~0. Impact: thousands of transitions. Format: Telegram/Discord group with automatic mentor-mentee matching by skills. Initial investment: one web page, a matching bot, and 50 emails to people in the Romanian tech community.

Second Foundation action: We identify 50 “influence nodes” from Romanian tech (CTOs, tech leads, community organizers), send them a personalized message with the data from this analysis, and ask them to mentor 5 people. Format: Twitter/LinkedIn thread + landing page. Budget: <500 EUR. Probability of reaching critical mass: 65%.

Seed 2: “The Living Catalogue” — Public database of successful transitions

The concept: A simple, open-source site where people who have successfully retrained post their story: what they did before, what they learned, how long it took, where they work now, what salary they earn. Transparent, optionally anonymous, verifiable.

The propagation mechanism: Each story posted inspires 10-20 visitors to begin their own transition. Each successful transition becomes a new story. Journalists pick up the most powerful stories → articles → more visitors → more stories. A positive feedback loop. After critical mass (~500 stories), the site becomes a national reference without any promotion.

Second Foundation action: We launch the catalogue with 20 initial stories (collected from our community). A Romanian tech influencer with 50K+ followers posts the first story. Format: viral video (60s) + link. Budget: 200 EUR production + 300 EUR boost. Probability of going viral: 45%. If it catches on, we reach 500 stories in 6 months organically.

Seed 3: “The Practice Company” — Real projects for retrained workers

The problem: Retrained workers cannot find jobs because they lack experience. They lack experience because nobody hires them. A vicious cycle.

The seed solution: NGOs and municipalities have dozens of unfinished digital projects (websites, reporting applications, form digitization). We connect teams of 3-5 retrained workers with these projects. They work for 4-8 weeks, deliver something real, and receive a reference and portfolio. Win-win: the NGO gets it for free, the retrained worker gets verifiable experience.

The propagation mechanism: Municipality X posts: “We digitized service Y with a team of retrained workers.” Municipality Z sees it and wants the same. Employed retrained workers tell their laid-off colleagues. Companies that see the portfolio begin hiring retrained workers → more people sign up → more projects get done. Tipping point: 10-15 successfully delivered projects. After that, demand exceeds supply and the system self-organizes.

Second Foundation action: We contact 5 small municipalities with stalled digital projects and 15 retrained workers from the Seed 1 network. We organize the first 4-week sprint. We document everything on video. Format: op-ed article in the local press + Twitter thread with results. Budget: 0 EUR (volunteer work). Probability of replication: 70%.

Seed 4: “The Resilience Index” — Public pressure through transparency

The concept: A public score, updated monthly, measuring how prepared Romania is for the AI transition. Components: retraining rate, investment in retraining, number of public programs, re-employment rate, comparison with Poland/Czech Republic/Estonia.

The propagation mechanism: Month 1 — we publish the index on olivLaw. Month 2 — the press picks it up (“Romania last in the EU on AI retraining”). Month 3 — politicians react (“we have a plan”). Month 4 — the opposition uses it in debates. Month 6 — the index is cited in EU reports. The index becomes a self-perpetuating public pressure tool — once created, no one can ignore it because any improvement or deterioration is publicly visible.

Second Foundation action: We publish the first “AI Resilience Index Romania” on olivLaw. We send it to 10 economic journalists (HotNews, Economedia, Profit.ro, Capital). Format: infographic + analysis article. Budget: 150 EUR design. Probability of media pickup: 55%. After pickup, the index self-sustains — each month is a new news cycle.

Seed 5: “The Seldon Hackathon” — Catalyst event with network effect

The concept: A 48-hour hackathon with a single theme: “Build a tool that helps someone find their next job after being laid off.” Participants: active developers + retrained workers + HR professionals + entrepreneurs. Prize: real implementation, not money.

The propagation mechanism: Out of 20 teams, 3-5 will produce viable tools (algorithmic matching, career simulators, mentoring platforms). The best ones launch publicly. Each tool becomes an independent seed with its own user base and its own propagation. A single weekend produces 3-5 self-sustaining initiatives. Each subsequent edition (quarterly) adds another 3-5. After one year: 15-20 active tools, each with hundreds or thousands of users.

Second Foundation action: We organize “Seldon Hackathon #1” in partnership with a tech hub in Cluj or Bucharest (Impact Hub, TechHub, etc.). Promotion: the existing tech community + LinkedIn + announcement on olivLaw. Budget: 500-1,000 EUR (venue + catering). Probability of producing at least 2 viable tools: 75%.

The network propagation model

The 5 seeds are not independent — they amplify one another:

• Seed 1 (Mentors) produces successful retrained workers → who become stories in Seed 2 (The Living Catalogue).
• Seed 2 (Stories) attracts people who sign up for Seed 3 (Real projects).
• Seed 3 (Delivered projects) generates data for Seed 4 (The Resilience Index).
• Seed 4 (Public pressure) produces the media attention that amplifies all other seeds.
• Seed 5 (Hackathon) produces tools that accelerate Seeds 1 and 3.

Monte Carlo estimate: If all 5 seeds are planted simultaneously (total cost: <2,500 EUR), the probability that at least 3 reach critical mass within 12 months is ~60%. The probability that the combined network produces over 5,000 successful transitions within 24 months: ~40%. Cost per successful transition: under 1 EUR (vs. 500-2,000 EUR for classic government programs). This is the power of network propagation: the initial investment is negligible, but the compound effect is massive.

“The Second Foundation does not build the future. It plants the conditions in which the future builds itself.”

Methodology

Data: Bureau of Labor Statistics, Eurostat, INS Romania, ANOFM, layoffs.fyi, Glassdoor, LinkedIn Economic Graph, McKinsey Global Institute reports (2024), OECD Employment Outlook 2025, Goldman Sachs “The Potentially Large Effects of AI on Economic Growth” (2024). Monte Carlo model: 10,000 iterations, 12 independent variables, horizon 2026-2030. MiroFish consensus: 8 autonomous agents, 3 deliberative rounds. Disclaimer: estimates for Romania are based on partial data — INS does not publish granular layoff data by tech sector. The figures are interpolations and should be treated as estimates, not certainties.