Part three · Predictions ·
Paid to Be Liable.
From 2027 to 2030, the best-paid human work will be the work an insurer is willing to cover. Then, around 2030, the test flips: the person an insurer will not cover is the one who refuses to use AI.
By John Rector · about 9 minutes
Earning power, as an insurer sees it
2026. Insurers are carving AI out of standard policies. The scarce human is the licensed one whose own coverage still reaches work done with delegates.
- Licensed and insurable, uses AI
- Licensed and insurable, refuses AI
- Uses AI, nobody liable
- Neither
Until 2030, income follows insurability. After 2030, insurability follows AI.
2027 to 2030
The scarce human is the one an insurer will cover for the work their delegates do. One signature now covers the output of what used to be a team.
From 2030
In field after field, AI-assisted work becomes the standard of care. The radiologist who reads alone becomes the one an insurer will not cover.
Monday, 7:40 a.m., 2028On Friday a structural engineer set the loads and limits for eleven projects. Her delegates worked under her direction all weekend. This morning she checks their calculations, corrects two of them, and seals four sets of drawings before lunch. Her seal is the job. Everything else came cheap.
The measure is a policy
In Not Moral. Not Responsible. Liable. I argued that commerce runs on liability, and that an idea that cannot be insured may as well not exist. This piece follows that to its practical end: which human jobs are worth the most in the next few years?
My answer is the jobs that someone will insure. A malpractice or errors-and-omissions policy is the market putting a price on your signature. It says: this person can be held liable, and here is how much liability we will stand behind. In 2027, that is the most honest measure of a human’s earning power. A résumé tells you what someone did. A policy tells you what they can answer for.
The timing matters. Through 2026, carriers have been writing generative AI out of standard liability policies, and the exclusions are spreading beyond general liability into directors and officers, employment practices, fiduciary, cyber, and errors and omissions coverage. So the scarce human is not merely the licensed one. It is the licensed one whose coverage still reaches work done with delegates.
A résumé tells you what someone did. A policy tells you what they can answer for.
The roles worth the most until 2030
They share three things: a license, a named person, and a policy behind the name. Delegates make each one more valuable, not less, because one liable signature can now cover far more work.
- Signers
- The engineer who seals drawings. The architect of record. The accountant who signs an audit opinion. The attorney of record. The physician of record.
- Qualifiers
- The licensed contractor whose license a company works under. The master electrician and master plumber. The home inspector. The appraiser.
- Fiduciaries
- The trustee. The executor. The registered investment adviser who owes a duty to the client, not to the firm.
- People of record
- The pharmacist in charge. The broker of record. The compliance officer. And a new one: the human of record who stands behind an AI-delivered service.
Current law already puts the weight on that person. In American medicine, physicians are held accountable for malpractice whether or not AI was involved. In engineering, a seal may go only on work done under the engineer’s responsible charge, which the national model law defines as direct control and personal supervision. Reviewing finished drawings after the fact does not qualify.
That last rule is the key to the whole period. Liability cannot be rubber-stamped. The person who carries it has to be directing the delegates from start to finish. That is what managing an AI looks like when someone can be sued over the result.
The flip: a tugboat without a radio
In March 1928, two tugs towing coal barges from Norfolk to New York ran into a gale off the New Jersey coast, and two barges sank. The tugs had no working radios, so they never heard the weather warnings that would have sent a prudent master to shelter. Radios were not yet standard equipment on tugs. The court held the owner liable anyway.
Judge Learned Hand wrote the line that has followed every new technology since: “a whole calling may have unduly lagged in the adoption of new and available devices.” Custom does not set the standard of care. If a precaution is cheap, reliable and available, failing to use it can be negligence, even if nobody else uses it either.
That is where radiology is heading. Legal commentators already call it inevitable that the models will outperform human readers and become the standard of care. Researchers are testing how mock jurors judge a radiologist when AI flagged something the radiologist missed, and the answers depend on exactly how the AI was used. Today the question is how a doctor used AI. Around 2030 it becomes whether they used it at all.
When that happens, the insurer’s question changes too. Before 2030 it asks: who is liable for the work your delegates do? After 2030 it asks: why are you not using them?
A renewal form, 2031Question 14: “List the AI systems used in your diagnostic workflow. If none, explain why.” A radiologist who has read films alone for thirty years leaves it blank. The quote that comes back carries a surcharge, and a note recommending a second-read system.
The only position that wins in both eras
Put the two halves together and the career advice is short. Before 2030, the scarce input is liability, so be a person an insurer will cover. After 2030, the required input is AI, so be a person who runs delegates well. The position that wins on both sides of the flip is the same: a licensed, insurable human who directs delegates and signs for what they produce.
The positions that lose are also clear. Work that nobody signs gets priced at the cost of a sentence, because delegates do it cheaply and nobody can be held to it. And the licensed professional who refuses AI keeps earning until about 2030, then finds the policy that made the license valuable getting harder to buy.
Five predictions
The best-paid new roles from 2027 to 2030 will be licensed signers who manage delegates.
One seal, one signature, one policy will cover the work that used to need a team. The value of each liable signature rises every year until 2030.
Proof of insurability becomes the résumé.
Customers and employers will ask what an insurer will write on you before they ask where you studied. The policy limit becomes a public measure of professional trust.
Insurers will reward AI use, then require it.
First discounts, then questions on applications, then surcharges for non-use, starting in imaging and diagnostics and reaching other licensed fields soon after 2030.
“Responsible charge” becomes the most important phrase in AI work.
Licensing boards will spell out how a human must direct delegates for a seal or signature to be valid, and will discipline people who rubber-stamp.
Unlicensed knowledge work loses pricing power fastest.
Work with no liable human behind it will be priced at what a delegate costs. The people doing it will either become people of record or compete with software.
What I would tell my students
My advice to the students I work with has always been to let your first employer be your first customer. Now I would add a second question to ask before any career move: what would an insurer write on me? If the answer is nothing, that is the gap to close. Get the license. Earn the right to sign. Learn to direct delegates the way the law already requires: in charge from start to finish. Do it now, while liability is the scarce thing, and you will already be the person the post-2030 standard is written around.
The claim, and how it fails
- Already true
- Physicians carry malpractice liability whether or not AI was involved. Engineers may seal only work under their responsible charge. Insurers are excluding AI from standard and professional policies. The Hooper rule says custom does not excuse ignoring an available safeguard.
- What has to happen
- Evidence keeps showing that AI-assisted work beats unassisted work in specific fields. Courts apply the Hooper logic to AI. Insurers build underwriting around how, and whether, professionals use it.
- Where I am probably wrong
- A large, correlated AI failure before 2030, one error repeated across thousands of reads or filings, would delay the flip. Insurers would punish AI use instead of its absence, and “human only” would sell at a premium for a few more years. And if boards cap how much work one signature can cover, the value of each seal stops climbing.
Monday, 7:40 a.m., 2031The same engineer, the same desk. Her renewal asked how she supervises her delegates, and she had a good answer. Down the hall, a colleague who swore he would never use them is filling out a form that asks why.
Until 2030, get paid to be liable. After 2030, you will be liable for not using it.
Background
- The T.J. Hooper, 60 F.2d 737 (2d Cir. 1932) (L. Hand, J.).
- NSPE, Responsible Charge position statement, citing the NCEES Model Law.
- FAQs About AI in Radiology: Legal Risks, Liability, and Malpractice, AJMC, March 2026.
- The social and legal costs of ignoring AI diagnostics, News-Medical, March 2026.
- Medical malpractice and AI: 4 response strategies when rejecting an algorithm’s advice backfires, Radiology Business.
- Insurers Start Excluding AI Risk in Some Policies, June 2026.
- We Don’t Generate Ideas. We Actualize Them. and Not Moral. Not Responsible. Liable.