What jobs are genuinely safe?

“Safe” isn't a useful frame on a 10-20 year horizon. ‘Currently resilient’ includes hands-on healthcare, skilled trades, regulated services with high accountability, deep relationship work, and creative work where taste and authorship matter. None are immune; all are slower to disrupt than routine knowledge work.

Should I learn AI to be AI-proof?

Yes, but not in the way most courses sell. You don't need to be an AI engineer. You need to be fluent at using AI tools well — knowing what they're good at, what they fail at, and how to specify what you want. That fluency is achieved through 50-100 hours of real use, not certification courses.

Are creative jobs done?

Some creative jobs at the routine end (stock photo, copywriting boilerplate, basic graphic design, simple illustration) are seeing serious substitution. Creative work driven by taste, authorship, brand, and human connection is doing fine. The line moves yearly; the principle holds.

What about new categories of work AI creates?

Genuinely new — AI-product builders, AI integration consultants, AI safety / governance professionals, AI-augmented specialists in every field. Most of these are early; the supply of credentialed practitioners is still small. Worth tracking even if you don't pivot.

Is being a generalist or a specialist safer?

Neither in isolation. The current evidence suggests ‘T-shaped’ or ‘comb-shaped’ profiles — deep expertise in 1-3 areas plus broad competence in adjacent fields — are most resilient. Pure generalists are easier to substitute; pure specialists are vulnerable to category disruption.

Should I worry about my kids' future jobs?

Yes, in the calibrated sense of helping them build durable underlying skills — writing, judgement, taste, learning-how-to-learn, social intelligence — rather than betting on specific job titles. The jobs they'll do at 30 may not exist at 16. The underlying capacities are evergreen.

AI-proof career

The ‘AI-proof’ myth

The framing implies there's a category of work AI will never touch. The honest version is that there's a continuum of substitutability, and that continuum keeps moving. Roles that looked safe in 2022 (copywriter, junior coder, basic illustrator) look much less safe in 2026. Roles that look safe in 2026 may look different in 2030.

What doesn't change is the underlying logic. AI is cheap, fast, scalable, and increasingly capable. Humans are expensive, slow, hard to scale, and increasingly capable at things AI is bad at. The durable career strategy is to be sharp at the things AI is bad at and to use AI well at the things it's good at.

Five properties of AI-resilient work

The work that's most defensible against substitution tends to share five properties. The more your role has, the more resilient it is.

Judgement under uncertainty. Decisions where the data is partial, the stakes matter, and someone needs to be accountable for the call. Most strategic work, most senior medicine, most senior legal, most leadership.
Trust-based relationships. Work whose value comes from long-term relationships that travel with the person, not the tool. Sales of complex products, long-term coaching, family-doctor relationships, therapist relationships.
Physical presence. Work that requires hands, bodies, locations — surgery, skilled trades, hospitality, performance, hands-on healthcare. Robotics is coming, but slower than language models.
Taste and authorship. Creative work where the value comes from a specific person's aesthetic, voice, or eye. Brand-building, certain kinds of design, certain kinds of writing, certain kinds of artistry.
Accountability. Work where someone's name has to be on the outcome — regulatory, fiduciary, life-and-death. AI can support; humans bear responsibility.

Where the resilient roles actually live

Rather than list job titles (which date quickly), here's the pattern:

Senior versions of most knowledge-work roles are more resilient than junior versions. The judgement layer outlives the execution layer.
Roles with regulated accountability (healthcare, law, engineering with PE, regulated finance) move more slowly because liability is hard to delegate.
Skilled trades that require hands, judgement, and on-site presence are highly resilient on a 10-year horizon, though some (basic plumbing diagnostics, certain electrical) will see augmentation.
Distribution-based work — anyone who owns an audience that travels with them — gains leverage rather than losing it as content production becomes cheaper.
Founder / operator / business-owner work is more resilient than employee work, on average, because the human is the accountable centre.

How to build toward AI resilience

Audit your current role. Use the AI career exposure audit. Find the routine percentage.
Move your hours toward judgement / relationship / authorship / accountability. Even within your current role, the mix is changeable over 6-12 months.
Use AI well in the parts you keep doing. The person who ships 3x with AI is more valuable than the one who refuses to use it.
Pick one durable skill to deepen. See high-income skills.
Build distribution — audience, network, reputation — that travels with you. Public output, deliberate network maintenance, one piece of visible work per year.
Build financial resilience. 6-12 months of cash buys you the freedom to make career decisions from strength, not panic.

Anti-patterns

The most common ways the “AI-proof” instinct goes wrong:

Refusing to use AI on principle. Reduces your productivity in the short term and your relevance in the medium term.
Pivoting wholesale every 6 months as new AI capabilities ship. Career strategy isn't day trading.
Buying every “AI-proof your career” course on offer. The course economy is monetising the anxiety. Most courses don't teach the durable underlying skills.
Going into a “safe” field you have no aptitude for. A job where you're mediocre is worse than a job where you're excellent and exposed.
Doomerism as identity. Becoming the loudest AI critic in the room doesn't protect your role; it gets you sidelined.

Common mistakes

Treating “AI-proof” as a real category.
Confusing AI tool fluency with AI engineering — you don't need to build the models to use them well.
Hiding from AI rather than learning where it helps you.
Underestimating how much your skill stack transfers.
Optimising for the next year and ignoring the next decade.
Not building any distribution of your own.
Letting anxiety drive the decisions.

Topic: AI career disruption.
Topic: High-income skills.
Topic: AI risk literacy.
Micro-course: Career Resilience in the Age of AI Disruption.
Tool: AI career exposure audit.
Path: AI-Era Personal Strategy.