David Arnoux recently shared something that caught my attention: "This headline isn’t actually about jobs.

It’s about a Nobel prize you haven’t heard of…

It’s a story about what happens when signals become free." When I read that, I immediately thought: this is the clearest explanation for why modern hiring can feel both faster and less trustworthy at the same time.

Arnoux’s point is not that AI is bad, or that automation should be rolled back. It’s more subtle. He’s describing a shift in the economics of credibility. When the cost of producing a signal collapses, the signal loses meaning. And if both sides of the hiring market rely on those signals, everyone ends up optimizing for noise.

The Nobel Prize idea hiding inside your inbox

Arnoux references economist Michael Spence’s work on signaling. The simple version: when two parties do not know each other, they use signals to infer quality. But credible signals usually have a cost.

A classic example (also echoed in Arnoux’s post) is the peacock’s tail: it is expensive to grow and maintain. That expense is the point. It signals fitness precisely because it is hard to fake.

In hiring, we have a whole ecosystem of signals:

Education credentials
Brand-name employers on a resume
Writing quality in a cover letter
Clean portfolios and polished LinkedIn profiles
Interview performance and preparedness

Some of these signals are imperfect, even unfair. But historically they had one redeeming property: producing them took time, effort, or opportunity cost. That cost created a kind of filter.

When signals become cheap, we get "cheap talk"

Arnoux puts it bluntly: "Signals only work when they cost something. When they’re free, economists call it "cheap talk." Cheap talk is worthless. Anyone can say anything."

That phrase, "anyone can say anything," is the heart of what many recruiters and hiring managers are experiencing right now.

Before generative AI:

A tailored resume implied time spent.
A strong cover letter implied thought and writing ability.
A thoughtful outreach message implied real intent.

After generative AI:

A tailored resume can be generated in seconds.
A cover letter might reflect the prompt more than the person.
Outreach can be mass-produced with personalization tokens.

None of this means every AI-assisted artifact is dishonest. Plenty of great candidates use AI like a spelling checker or brainstorming partner. The issue is that, at scale, the market can no longer distinguish effort from automation. When the cost approaches zero, the signal stops separating high-intent from low-intent.

"It drove the cost of nearly every signal to zero." - David Arnoux

The hiring version of "two bots talking"

Arnoux also points out the second half of the loop: companies started automating judgment.

AI generates the resume, bullet points, and cover letter.
AI screeners parse and score that content.
Automated workflows advance or reject candidates.

In other words, you can end up with two systems optimizing against each other. One system learns to produce text that matches patterns. The other learns to reward text that matches patterns. Neither is necessarily connected to job performance.

This explains why so many people report the same odd experience:

Highly qualified candidates get rejected instantly.
Underqualified candidates get through with polished materials.
The process feels efficient but strangely detached.

Efficiency rises. Trust falls.

The Tinder analogy: engagement is not connection

Arnoux mentions The Atlantic’s comparison to Tinder and dating apps. The analogy holds because both markets depend on trust under uncertainty.

On dating apps, profiles are signals. When filters, templates, and optimization techniques get too good, the profile becomes less predictive of the real person. At the same time, the platform can optimize for the wrong objective (more swipes, more time in app) rather than durable matches.

In hiring, the wrong objective can look like:

Minimizing time-to-review rather than maximizing quality-of-evaluation
Maximizing applicant volume rather than maximizing fit
Automating rejection at scale rather than improving calibration

You can end up with movement but not progress. Lots of activity, little belief.

The uncomfortable takeaway: friction is the mechanism

Here is the part of Arnoux’s post that I keep coming back to: "The strange implication: the solution isn’t efficiency. It’s friction." That sounds counterintuitive in a world that praises speed.

But if you accept signaling theory, it makes sense. Signals become trustworthy when they impose a cost that low-quality or bad-faith actors cannot easily pay.

That cost does not have to be money. It can be:

Time
Personal accountability
Public reputation
Specific effort tied to the role

Arnoux’s examples are simple and powerful:

Referrals work because someone risks their reputation.
Handwritten notes work because they take time.
Long-form work samples work because they’re hard to fake.

These are not perfect either. Referrals can be biased. Work samples can disadvantage people with less free time. But they share something that generic AI-generated text does not: commitment you cannot fully automate.

What credible signals look like in an AI-saturated hiring market

If I were advising a hiring team using Arnoux’s framework, I would focus on signals that are expensive in the right way: costly enough to be meaningful, but not so costly that you only select for people with excess privilege.

For companies: redesign the funnel around reality, not polish

Ask for work that is hard to bluff

A short role-specific exercise (60-90 minutes max)
A critique of a real artifact (a landing page, a sales email, a support macro)
A mini plan (first 30 days, key risks, what they would measure)

Introduce accountable human judgment

A hiring manager spends real time reading a small set of candidates deeply
Structured interviews with defined scoring rubrics
Calibration sessions to reduce "vibes" decision-making

Use AI, but move it upstream and downstream carefully

Upstream: let AI help craft better job descriptions and outreach, but keep them honest and specific.
Midstream: avoid treating AI scoring as truth. If you use it, treat it like a triage assistant with audits.
Downstream: use AI to summarize interview notes, not to replace the interview.

For candidates: compete on proof, not prose

If resumes and cover letters are cheap talk now, your advantage is evidence.

Build a portfolio of real work (even small)
Write one or two deep case studies: context, constraints, tradeoffs, results
Share your thinking publicly: a short essay, teardown, or analysis
Get a referral when it is authentic, and make it easy for the referrer to be specific

The goal is not to look impressive. It is to make it easy for someone to believe you.

As David Arnoux mentioned, "The things that can’t scale are the things that still mean something."

The paradox: efficiency has become the enemy of trust

Arnoux ends with a line that deserves to be a hiring mantra: "As AI makes communication free, the only credible signals will be the ones that remain expensive. Human judgment. Personal risk. Time that can’t be recovered."

I read that as a challenge to both sides of the market:

If you are hiring, stop worshipping throughput. Protect trust.
If you are applying, stop optimizing for keyword match alone. Create proof that survives skepticism.

Friction is not automatically good. But the right friction produces clarity. It makes room for signals that are anchored in reality.

In an era where anyone can generate a perfect paragraph, the competitive edge is being able to show your work, stand behind it, and invest time in what matters.

This blog post expands on a viral LinkedIn post by David Arnoux, Helping GTM Leaders & Founders Grow With GTM x AI | Fractional CxO | Building Linkedin Tools @ humanoidz.ai. View the original LinkedIn post →