The Pragmatic Engineer just described the problem we’ve spent seven years building for

Gergely Orosz just published a three-part investigation into the 2026 tech job market (Part 1, Part 2, Part 3). He collected more than fifty firsthand accounts from hiring managers, recruiters, founders, and engineers.

Part 3 describes a market where nobody can find each other, and nobody trusts what they find. One of his sources, a Big Tech program manager, put it plainly: your resume doesn’t get seen “unless you know people who can vouch for you.”

Seven years ago, we had to explain what vouching meant in a hiring context. Now the word shows up in the biggest engineering newsletter in the world. As the diagnosis.

The catch-22 his sources describe

The same pattern repeats across his anecdotes.

Companies can’t fill certain roles. Product-minded engineers. Devtools and infrastructure people. Engineering managers and Staff+ engineers. Some companies can’t close these candidates even at top-of-market pay.

Meanwhile, those same companies have stopped reading applications. One CEO told Gergely he gets about a thousand a day. Roughly two are relevant. His team now hires only through their network and outbound sourcing.

And the engineers who fit those hard-to-fill roles? They’re applying. Into the pile nobody reads.

So the companies never find the engineers. The engineers never reach the companies. Both sides think the market is broken. Both sides are right.

Why this happened: the signal died

For most of hiring’s history, an application meant something because it cost something. Time. Effort. Attention.

AI dropped that cost to zero. And when a signal is free to produce, it stops carrying information.

The rest follows. Employers can’t tell good applications from bad, so they stop reading. Good candidates hear nothing, so they stop applying. What’s left in the pile is exactly what employers feared. Economists call this a market for lemons. Hiring just became one.

So trust moved to the last signal that still costs something to fake: a real person putting their reputation on the line for you. That’s why “referrals” is the answer in nearly every one of Gergely’s fifty anecdotes. Referrals didn’t get better. Everything else got worse.

But referrals have a flaw. They only travel inside networks that already exist. The engineers stuck in the pile aren’t unqualified. They’re un-networked. Those are different problems, and they need different fixes.

We bet on this moment years ago, when the bet looked strange. For most of those years, being early looked exactly like being wrong. It doesn’t anymore.

Part 3 doesn’t describe noise. It describes fraud.

The usual take is “hiring is hard right now.” What Gergely documents is worse than hard. It’s deliberate deception, at every stage of the funnel.

AI-polished resumes that don’t match the person who shows up. Resumes stuffed with AI keywords to sell mid-level experience at senior prices. Cover letters nobody reads, because everyone knows a model wrote them. Interviews where an AI feeds answers from a second screen. Interviews taken by a different, better person than the one who starts the job. And at the far end: fully invented candidates, including suspected North Korean operatives targeting remote roles at Western companies.

That last one matters more than its frequency suggests. When you can’t tell a real remote engineer from a fake one, the safe move is to stop hiring remote at all. Gergely’s data shows companies doing exactly that. The fraud tax lands on honest engineers everywhere outside the big hubs.

What we built: two layers, because one isn’t enough

Telescoped has two layers. Part 3 shows why we need both.

The first is the recommendation graph. Verified recommendations from senior engineers who actually worked with you. Every one is a named, real person staking their own reputation on yours. This is what Gergely’s source meant by people who can vouch for you — made portable, so it travels past your first-degree network to the companies actually searching.

The second is a live interview on the platform. An agent asks real questions and pushes hard on the answers. We spent a year scoring real sessions by hand, transcript by transcript, to learn what separates real expertise from a good performance.

Three things we learned.

Long answers mean nothing. In our sessions, the longest answers were the most likely to contain pasted or AI-polished text. The most authentic candidates were often terse. Any filter that rewards volume now rewards fraud.

The real tell is what happens under follow-up. Pasted fluency can’t survive “what was the exact failure mode?” A polished first answer followed by a cliff is the clearest fraud signature we found. We call it paste-then-collapse.

And gaming isn’t hypothetical. We’ve already caught candidates feeding pre-written scripts into the system to build a persona. The fraud showed up in our data before we went looking for it.

The map: every fraud vector in Part 3, against the trust stack

What Gergely documents	Why it beats resume screening	How the trust stack answers it
AI-polished resumes that don’t match the person	Screening judges the document. AI perfected the document.	Certification never reads the resume. A live session tests the person. And recommendations come from people who watched the work happen.
Keyword stuffing to inflate seniority	Keyword filters reward vocabulary. Vocabulary is now free.	We score specifics, not terms. Follow-up questions collapse borrowed expertise. Length earns nothing.
AI-generated cover letters	Self-written prose carries no information anymore.	We give it zero weight unless you have been recommended. What counts: recommendations, live answers, and a real person’s staked name.
AI feeding answers during interviews	A human interviewer can’t tell AI latency from thinking.	Timing tells the truth. A formatted 300-word answer, fourteen seconds after the question, fails the math no matter how good it sounds. Then follow-up exposes the collapse.
Interviews outsourced to a different person	One interview verifies a performance, not a person.	Scores build across sessions, over time. And every recommendation is a named engineer you can call, vouching for this specific human.
Fully fabricated candidates	A fake candidate costs a resume generator and a Zoom account.	Here, a fake has to survive live probing and convince multiple real engineers to lie in public, in a system that remembers.

One pattern runs through all six rows. We don’t try to make fraud impossible. We make it expensive. Today, a fake senior engineer costs almost nothing to run. Against staked reputation and live probing, the same fake costs more than the job pays. That’s the whole design.

What this doesn’t solve

Honesty cuts both ways.

Portable vouching doesn’t fix employer ghosting, which Gergely shows flowing in both directions. It doesn’t create demand for roles the market has genuinely cooled on. And it can’t help someone who wants trust without earning it. The certification bar is one you can’t paste your way over.

What it does fix is the routing failure at the center of the catch-22. Companies get a channel where every candidate arrives pre-trusted. Skilled but un-networked engineers get a way to be found.

Where this goes

Gergely ends by noting that personal networks matter more than ever. We’d sharpen that. Staked reputation matters more than ever. Networks are just where it used to live. The job now is making it portable.

If you’re a hiring manager who stopped reading inbound: that was the rational move. But closing a broken channel isn’t the same as giving up on discovery. Talk to us about what a vouched, certified pipeline looks like.

If you’re an engineer applying into the void: you’re probably not the problem. You’re un-networked, not unqualified. Here’s the on-ramp.

And to Gergely and the fifty-plus people who talked to him: thank you. It’s easier to fix a problem once the best analyst in the industry has proved it exists.