Back to Devoured - November 16, 2026
DEVOURED
AI and hiring: bad for employees, bad for employers

AI and hiring: bad for employees, bad for employers

Tech John Hempton
Generative AI is breaking public-facing processes like hiring and government complaint handling by making it trivial to generate cogent but potentially false submissions.
What: John Hempton, founder of Bronte Capital, reports that AI-generated job applications have forced him to abandon personalized interview processes. Simultaneously, the Australian Ombudsman's office is struggling to triage complaints as AI-assisted submissions mask non-substantive or 'hallucinated' claims, removing the effective filter of requiring clear, human-authored explanations.
Why it matters: This reveals a systemic 'tragedy of the commons' where the low cost of high-quality AI text generation destroys the signaling value of public input, rendering traditional human-centric intake systems unscalable.
Decoder
  • Hallucination: The phenomenon where an AI generates text that sounds plausible and authoritative but is factually incorrect or nonsensical.
Original article

AI and hiring: bad for employees, bad for employers

AI breaks things - and there is no obvious way to put them back together

If you talk to young people you will find how AI is disrupting the job market - especially at graduate entry level. Many are (justifiably) distressed about their prospects.

I think we can now safely say that AI is bad for entry-level employees in most intellectual activities.

I want to talk about this from the other side - how AI is disrupting the employer’s prospects of getting good staff.

At Bronte (the financial firm I founded with friends) we hire left-field candidates. We do not seek the usual financial market credentialism (I have always regarded a CFA as a mild negative on a CV).

We sought raw intelligence and curiosity and that was hard to screen for by reading CVs.

We had a policy explained to applicants. We interviewed everyone who sent us a CV in any way personalised to us and with any reasonable claim to a position.

We sent applicants this form letter:

We interview far more widely than most firms - and we interview a large number of people.

The problem is that if you interview more people you also have to interview quickly. We eliminate quickly.

This means that most candidates will get a five-minute interview. We can appear brusque when we cut the interview short - and we are sorry about that - but it is from our perspective the necessary part of interviewing quickly.

There are three reasons this interview might last more than ten minutes.

a. We are genuinely interested in hiring you

b. We are learning something - that is by listening to you talking about what you are doing we might get better at what we are doing. We may still be interested in hiring you.

c. You are a good-looking contra: you are pitching us a long and we are salivating about it as a short for instance. We can make money off misconceptions you bring to us.

All three of these have happened - and at the end of a longer interview we will discuss with you which category you fit in.

If it is c (as above) we would appreciate a follow-up chat in a few years - and one of us will be wrong. If it is us that were wrong we want to improve by learning what mistakes we may have arrogantly made. If it is you - you should follow up too - and see what you can learn.

It can very easily be either of us that is wrong.

I know that c (above) looks exploitative - we are essentially leading you on. It is not totally exploitative. We may still be interested in hiring you two years later when you have made us look like fools.

Having run a couple of hundred such interviews over the years we have hired a handful, and found a half dozen more jobs at other firms for which they were better fits. [Nobody in either camp has reason to complain.]

We have also found a dozen or so shorts (and noted which firms the bozo idiots who pitched us frauds wound up at).

We have however been brusque a few times - and many applicants have found the quick rejection difficult (and some have told me so).

Alas none of this works any more. We used to run 40-60 job interviews a year - and 5-10 minutes once a week was a tolerable use of my time. Especially if I learned something.

Generative AI now means that I get five times as many applicants. And I can’t run 300 job interviews a year or I would do nothing.

So I have to cull. And I do.

I could get AI to select which of the hundreds of applicants to choose - but AI picking between AI generated applications is worse at selection than a human picking through human generated ones.

Now having made what is probably a bad cull I get applicants on the phone for a five minute chat. And sometimes I get really good answers to hard questions. Nuanced and clever.

I have thought some applicants were astonishing on the phone. And then I got them in for what might normally be three or four hours of chatting. And they were decidedly ordinary - and should never have got through the first cut.

We think they had AI help on the phone.

From our perspective this is a mess.

The problem is widespread

I was chatting to a senior person at a big consulting firm about this. This was a firm that did more than a bit of credentialism (hiring heavily from target universities like Harvard).

They used to get 3-5000 applicants each year. They would winnow this down to 300 and then give them an extensive workshop style interview with case studies and the like. From that they might choose 80 to hire.

They had a really big problem - which is that they could not work out how to winnow down from 3000 to 300. The applicants all looked good - but they also looked similar. Their process has broken too.

AI is going to break many things

I was chatting to friends in Canberra (the political capital of Australia) about this. They told me that AI is breaking the Ombudsman’s office.

The Ombudsman is a government agency where you can - at first instance - send complaints about the behaviour or decisions of other Government departments. This often works - in Australia. The Ombudsman investigates and what is really a simple problem gets resolved quickly and relatively cheaply.

The Ombudsman has always had more complaints than they can investigate thoroughly. And they had a solution. They would triage to prioritise anyone who could write a 3-5 page crystal clear explanation of their issue. If the applicant could explain the problem clearly then the problem was probably real and worth investigating.

This worked. Simply writing out the problem clearly removed the (many) crackpots who sent complaints.

Alas this no longer works. Any AI can make a complaint sound cogent (even when it is not). It can hallucinate cogency (especially when prompted to do so). The Ombudsman’s office is rapidly breaking.

This is not good. I think democracy works better if you have a rapid, cheap and effective way of complaining about government actions. And one of those is being removed.

A general rule to what breaks

Anywhere where you seek input from the public, whether it be a job application process, or an Ombudsman’s complaint process, or just social media will have a tendency to break.

And I am not sure how we solve this problem.

I have toyed with requiring job applicants to hand-write a one page cover note. But that is a solution out of desperation.

I am hoping people have something better.

Thoughts?

Explore all 69 articles in today's digest

Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi