September 22, 2026 · 9 min read · Editorial ranking, framework disclosed

Jobs AI Can't Replace: The 2026 List, Ranked by What Actually Protects Them

Every list like this ranks by vibes. This one ranks by a stated key - how many of four durable properties each career stacks - and shows you the key, so you can disagree with the ranking on the merits. It also carries the column these lists always leave out: what would move each career down.

How this list is ranked

By moats stacked. The Four Moats are the four properties that keep showing up in work AI struggles to absorb - judgment under ambiguity, trust-based relationships, hands-on skilled work, and accountability for outcomes. A career that stacks three of them is in a stronger position than one that stacks two, because losing a moat still leaves something underneath. The framework, with each moat explained in full, lives on our AI-proof careers pillar.

What the ranking is not: it is not an output of a dataset. We measure how threatened people feel, not how exposed their tasks are, so the mapping below is an editorial read through a disclosed framework. Within a tier we make no ordering claim at all. And no salary or growth figures appear anywhere on this page - we publish numbers we can point at, and we do not have verified ones for these occupations.

The ranking

Tier 1 - three moats deep

These stack three of the four. That is not a promise of safety; it is the widest gap between what the work requires and what a model can supply on its own. Note how many involve a body in a room and a name on an outcome at the same time.

#	Career	Moats stacked	What would move it down
1	Nurse practitioner	Trust · Hands-on · Accountability	A seat that is mostly protocol-following and charting, with the diagnostic calls made above you
2	Paramedic / emergency responder	Judgment · Hands-on · Accountability	Very little inside the role; the realistic risks here are staffing and pay, not automation
3	Consultant / fractional executive	Judgment · Accountability · Trust	Selling deliverables rather than decisions - the deck is the part that automates
4	Construction / trades supervisor	Judgment · Hands-on · Accountability	A coordinator seat that is mostly scheduling and status reporting with no authority over the work

Tier 2 - two moats deep

Two moats is still a strong position - most work in the economy stacks zero or one. What separates the careers here is that the second moat is doing real load-bearing work, so losing one still leaves something defensible underneath.

#	Career	Moats stacked	What would move it down
5	Electrician	Hands-on · Judgment	Repetitive new-build install work on standardized plans, where the diagnosis has already been done
6	Therapist / counselor	Trust · Judgment	Volume-capped work delivered through a platform that owns the client relationship, not you
7	Physical therapist	Hands-on · Trust	Handing off assessment and running prescribed plans someone else wrote
8	HVAC technician	Hands-on · Judgment	Swap-and-replace call queues where the diagnostic step is the software's, not yours
9	Veterinarian / vet tech	Hands-on · Trust	Corporate-chain seats built for throughput, where you meet the animal and never the owner
10	Complex / enterprise sales	Trust · Judgment	Moving down-market into transactional, single-decision-maker deals that self-serve
11	Teacher	Trust · Judgment	Delivering a scripted curriculum to a screen-based cohort you never build a relationship with
12	Field / site engineer	Hands-on · Judgment	Drifting into remote monitoring and report production, off the site and away from the call
13	Business owner / independent operator	Accountability · Judgment	A business whose product is itself a commodity output that customers can now generate themselves
14	Executive coach / facilitator	Trust · Judgment	Packaged content delivery - the curriculum sells, the person in the room does not
15	Senior operations leader	Judgment · Accountability	A seat where the decisions are made above you and your week is reporting on them

What AI is already absorbing inside each of these roles - the task layer, which is the half most lists never print - is on the companion table at AI-proof careers. Read the two together: this page is the ranking and the downside, that page is the framework and the worry data.

The four moats, in one table

The ranking key, stated once so you can apply it to a career that is not on the list - yours, for instance:

The moat	Why AI struggles with it
Judgment under ambiguity	Deciding with incomplete information, owning trade-offs, being accountable when the call is wrong.
Trust-based relationships	Work where the relationship is the product - where people need a person they trust, not an answer.
Hands-on skilled work	Physical skill in unpredictable environments - the frontier AI reaches last, because the world is messier than text.
Accountability for outcomes	Roles where someone must own the result - legally, financially, or reputationally. Software can draft; it cannot be responsible.

Four things people expect to see on this list, and why they are not moats

“Creative” is not a moat. Creative production is among the fastest-automating categories there is. Creative judgment - knowing which idea is worth making, and being accountable for the call - is judgment, which is already on the list under its real name.

“Technical” is not a moat. Coding is automating at the task layer while architecture and system judgment hold. A technical title tells you almost nothing about which layer your week sits in.

A credential is not a moat, but a liability often is. The degree does not protect you; the fact that someone must sign, and can be sued or struck off for signing wrongly, does. That is the accountability moat, and it explains why licensed roles keep appearing near the top without any claim that licensed people are more skilled.

Seniority is not a moat either - except where it changes the work. A senior title that mostly produces reports is a task-shaped seat with a good salary attached. A junior seat where you own a client relationship end to end has a real moat. The property is in the work, not the level.

The number that should calm the panic slightly

Worth holding against the headlines: across the 27,366 working adults who have completed our career assessment, about 1 in 10 say AI threatens their current job. And the professions punditry calls doomed are not the ones reporting the most concern - in our data the groups closest to the tools tend to report less worry than the groups furthest from them. That is a measure of how threatened people feel, not a measure of what will happen; we are careful about the difference. The full by-profession breakdown, with the counts and the date it was computed, sits on the pillar page.

The list is the wrong unit. Your week is the right one.

Here is the uncomfortable part of ranking careers at all: protection is a property of the seat, not the title. The same job appears in one organization as a three-moat role and in another as a task queue with a nice name. That is why the erosion column above matters more than the ranking - every one of those conditions is a real seat that exists today, inside a career on this list.

So the useful move is not to pick a job off this page. It is to work out how many moats your current week actually stands on, and whether the durable version of your career is one step sideways rather than a restart. If you want that done question by question, the self-assessment companion to this page is our 12-question check on whether your job is safe from AI, scored in your browser.

FAQ

What jobs can AI not replace?

Rather than a list to memorize, use the property that does the protecting. Work resists automation when it stacks what we call the Four Moats: judgment under ambiguity, trust-based relationships, hands-on skilled work, and accountability for outcomes. The careers that rank highest stack three of the four at once - nurse practitioners, paramedics, construction and trades supervisors, consultants and fractional executives. Careers stacking two follow: electricians, therapists, physical therapists, HVAC technicians, veterinarians, complex sales, teachers, field engineers, business owners, executive coaches, senior operations leaders. This ranking is our editorial read through that framework, not an output of a dataset - and the ordering matters less than the reason behind it, because the same job title can be a three-moat seat in one organization and a one-moat seat in another.

What makes a job hard for AI to replace?

Four properties, and they cut across industries rather than tracking them. Judgment under ambiguity: deciding with incomplete information and owning the trade-off. Trust-based relationships: work where the relationship itself is the product. Hands-on skilled work: physical skill in unpredictable environments, which is the frontier AI reaches last because the world is messier than text. Accountability for outcomes: roles where someone must be responsible, legally or financially or reputationally - software can draft, but it cannot be held to account. Notice what is not on that list: 'creative' and 'technical' are not moats. Creative production automates faster than creative judgment, and coding is automating at the task layer while system judgment holds.

Are these rankings based on data?

The ranking is editorial and labelled as such. It is our read of which careers stack which moats - we do not have a dataset that measures task-level automation exposure, and we will not dress an opinion up as one. What we do have is the worry side, from our own assessments: across the working adults who completed our career assessment, about 1 in 10 say AI threatens their current job, and the professions the headlines call doomed are not the ones reporting the most concern. That by-profession breakdown, with the counts and the computation date, is on our AI-proof careers page.

Should I change careers because of AI?

Usually no - usually you should change your seat, not your field. The useful audit is your own week: sort your hours into processing, drafting, scheduling and routine analysis on one side, and deciding, persuading, building trust and owning outcomes on the other. That ratio tells you more than any occupation-level forecast, because the ratio is what actually differs between two people with the same job title. A real field change earns its cost when your field is task-shaped from top to bottom - and then the target is one of the four moats rather than whatever is trending.

Which jobs will AI replace first?

Task-shaped roles: mostly processing, drafting, scheduling and routine analysis, where the output is checkable and an error is cheap. The framing that holds up better than any list is that jobs rarely disappear whole - tasks do. A role that is 80% task-shaped shrinks toward its judgment core; a role that is 80% judgment-shaped absorbs the tools and speeds up. This is also why 'is my job on the safe list' is the wrong question and 'what share of my week is task-shaped' is the right one.

Are the trades really safe from AI?

Hands-on skilled work in unpredictable environments is the hardest moat to cross, so the trades rank well here for a real reason. But read the erosion column rather than the headline: the protected part is the diagnosis and the site-specific problem solving, not the install. Repetitive new-build work on standardized plans, where someone else already did the diagnosis, is a weaker position than the same trade doing service and repair. The moat is the judgment inside the physical work, not the physicality alone.

Which of these is actually available to you?

A ranked list cannot tell you that - it does not know your experience, your constraints, or what you need to earn. Our free assessment reads those in your own words and returns three named paths with honest income ranges, built from the career capital you already have rather than a restart. Free in full.

Take the free assessment