I want to write something useful about AI and jobs. The discourse is so polarized that most articles are either alarmist (AI will take all the jobs, panic) or dismissive (AI just creates new jobs, relax). Neither is true. The actual answer is more interesting and more actionable.

Here's the honest version, with the numbers and the moves.

What's actually happening (the numbers)

Two data points that ground the conversation:

In April 2026 alone, 26% of US job cuts were AI-attributed. That's per the Challenger, Gray & Christmas monthly report, the standard layoff tracker. AI-attributed cuts have been accelerating: 22,000+ workers impacted in the first quarter, with the technology sector taking the largest share at 33,361 cuts in April alone.

Between January and June 2025, 77,999 tech job cuts were connected to AI adoption. That's hundreds of people losing their job every day from a single cause across a six-month window. The trajectory in 2026 is steeper.

Those numbers are real and they matter for the people inside them. But they're also a specific story. They're not the whole story.

The whole story is more like this: by the end of 2026, an estimated 85 million jobs will have been displaced by automation broadly (AI is a major contributor). By 2030, that number is projected to reach 92 million displaced. But — and this is the part the doom-headlines leave out — the same projections see 170 million new jobs created by 2030. Net positive. Net new jobs.

The catch in that net number is the word "net." The 92 million people whose jobs are displaced are not the same 170 million people who get the new jobs. The transition is what gets people hurt, even when the aggregate math works out.

Who's actually getting hit (specifics matter)

Headlines say "AI is coming for white-collar jobs." That's too broad. The specifics:

• Data entry: 95% of tasks automatable. This is mostly already gone. If you have a job that's "I take information from one system and put it in another," that role is going to be substantially or entirely automated within 18 months. The training cost is low. The error rate of AI on these tasks is now lower than human error rate. The economics close.

• Customer service representatives: 80% of tasks automatable. Most customer service inquiries are variations of a finite set of questions that map cleanly to a knowledge base. AI handles those well. The remaining 20% — genuinely novel issues, emotional escalation, complex multi-system situations — still needs humans, but you need fewer of them. Expect the role to compress to "tier-2 specialist" rather than "front-line representative."

• Administrative roles: 26% of all cuts. This is the largest segment. Calendar scheduling, document drafting, data summarization, meeting notes — the load-bearing parts of a lot of admin work are now AI-automatable.

• Production work: 13% of cuts. Different mechanism — physical automation, often with AI for QC and routing. Slower to deploy because capex-heavy, but the trajectory is the same direction.

• Tech: the largest absolute numbers. Junior engineering roles where the work is "implement this well-specified ticket" are the most exposed. Senior roles where the work is "figure out what to build and why" are the least exposed. The middle is where the squeeze lives.

The roles that are NOT under pressure in 2026: skilled trades (electrician, plumber, welder), healthcare delivery (nursing, physical therapy, surgery), early-childhood education, complex sales, executive leadership, and most creative work where the "creative" part is genuine taste rather than execution.

What it means for "knowledge workers"

If you do desk work, here's the realistic picture for the next 3-5 years.

The dichotomy is not "your job is safe" versus "your job is gone." The dichotomy is "you're augmented by AI" versus "you're replaced by someone who is."

Augmented: the workers using AI tools effectively become ~26% more productive within weeks of adoption. Across enough roles, that means companies need ~25% fewer workers to do the same work. But the workers who remain are more valuable. Recent data shows workers with advanced AI skills earn 56% more than peers in the same role without those skills.

Replaced: the workers who don't adopt AI tools are competing against augmented peers for shrinking headcount. Their relative productivity hasn't dropped — but the goalpost moved. What used to be "competent at your job" is now "competent at your job minus 26%." That's the gap that becomes a layoff over time.

The hard part is that this isn't a moral failing of the people who get displaced. It's a labor-market structural problem. The transition takes years. The new roles created don't appear in the same time, place, or shape as the old roles displaced.

What to actually do

I'm going to be specific. There are three moves, and they compound.

Move one: get fluent with AI tools in your actual workflow

Not "learn AI" in the abstract. Not a course. Specifically: pick one AI tool that fits something you do every week, and use it deliberately for that thing until you're faster with it than without it.

For a marketer: Claude for writing first drafts that you then edit, not for replacing your editing judgment.

For an engineer: Cursor or Claude Code for boilerplate and refactor work, with you still owning architecture.

For an analyst: ChatGPT or Claude for restructuring data and exploring hypotheses, with you still owning interpretation.

For an administrator: Calendar AI, drafting AI, summarization AI. Stack them. Your throughput should double in 60 days. Document the wins. Make them visible.

The pattern: AI as the leverage, you as the judgment layer. The judgment layer is where the wage premium lives.

Move two: shift your role description toward the judgment-layer parts

Most jobs have a mix of judgment work (decide what to build, evaluate quality, negotiate with stakeholders, set strategy) and execution work (write the code, draft the email, prepare the deck). AI eats execution faster than judgment.

If your role description is mostly execution work, get more of the judgment work. This usually means: take on the parts of your manager's job that they're happy to delegate. Make decisions instead of staffing them up. Own outcomes instead of completing tasks. The "you executing" parts of your job become smaller. The "you deciding" parts get larger. Your career durability follows the second curve.

Move three: develop one skill AI is genuinely bad at

The skills AI is bad at in 2026: complex interpersonal judgment, genuine creative taste (not "generate options," but "pick the right one and articulate why"), physical-world coordination, deep domain expertise built over years, and trust-based relationships.

You probably have one of these latent. Develop it deliberately. Make it visible in your role. When the layoff calculus runs, you want to be the person who's irreplaceable for a specific reason, not the person who's interchangeable with anyone else who does similar tasks.

What I think the next five years actually look like

A bumpy redistribution. Not "AI replaces everyone." Not "everyone retrains as a prompt engineer." Specific:

• Many existing jobs compress: same work, fewer people doing it, augmented by AI. This is most jobs.
• Some categories of jobs largely disappear: data entry, low-tier customer service, basic content generation. This is fewer jobs than the headlines suggest, but real.
• New categories emerge: AI orchestration, AI safety/audit roles, AI-assisted specialists (lawyers, doctors, engineers who use AI deeply). These appear gradually.
• The transition hurts. Specifically: people whose roles compress can keep working but earn less. People whose roles disappear face career-restart costs that the market doesn't currently absorb gracefully.

If I had to pick the single number that captures the situation: the workers with advanced AI skills earn 56% more than peers without them. That premium will keep growing for the next 3-5 years before it stabilizes. The decision to develop those skills now versus in 2028 versus not at all is one of the higher-stakes career decisions you'll make this decade.

What I'm not saying

I'm not saying "everyone will be fine." I'm saying the people most likely to be fine are the ones who treat this as a serious career planning issue right now, while there's still time and runway. The people who treat it as "AI will pass me by" are the ones at risk.

Also: this is genuinely a policy issue, not just an individual issue. The labor market is not going to absorb 92 million displaced workers smoothly through individual upskilling. That's a societal-scale problem that needs societal-scale responses. But individual-scale responses are still worth making, because they protect you regardless of how policy plays out.

Further reading

• SQ Magazine — AI Job Loss Statistics 2026 — comprehensive statistics roundup, sector by sector
• CBS News — AI accounts for 26% of April job cuts — current US layoff data
• World Economic Forum — AI's $15 trillion prize, won by learning — skills-development framing
• IMF — New Skills and AI reshaping the future of work — macro economic view
• McKinsey — Agents, robots, and us: Skill partnerships — augmentation-vs-displacement framing