HN "Who Is Hiring" Just Flipped: 2.2 Candidates Per Role

For a decade, Hacker News's monthly "Who is hiring?" thread was the cleanest signal in tech: more companies than candidates, every month, no exceptions. That broke in 2026. By September the "Who wants to be hired?" thread carried 2.2 candidates for every company posting a role, and the first week of October ran at 2.6.

If you are an engineer refreshing those threads on the first of the month, the arithmetic has already changed under you. This is what the flip actually looks like, why the headline ratio understates the pain for juniors, and the specific moves that still work when supply has doubled against demand.

What the HN threads actually show

Company posts on HN's "Who is hiring?" fell about 33% year over year while candidate posts rose 44% in six months, flipping the ratio from 1.2 to 2.2 candidates per hiring company between April and September 2026. The underlying dataset is 13 "Who is hiring?" threads from October 2025 through October 2026 (9,194 roles across 3,909 company posts) and 7 "Who wants to be hired?" threads from April through October 2026 (3,586 candidates), parsed into a fixed schema by an independent analysis on dev.to.

The movement is one-sided in a specific way. Each company still lists about 2.3 roles per post, unchanged from a year ago. So the decline is not shrinking job counts inside a stable set of employers. It is fewer employers, period.

That is a roughly 83% worsening in the supply-to-demand ratio over five months, on the single corner of the tech market that was supposed to be insulated.

Why the 2.2 ratio understates the pain

The HN headline ratio is a best case, not a representative one, because the denominator is self-selected toward profitable startups that still see HN as a signal channel. Broader tech is harder.

Look at who still posted in October 2026: Ramp, Sierra, Fal, Vanta, Framer, ElevenLabs. All AI-adjacent or infrastructure. None of them are hiring a generalist Rails developer out of a bootcamp. The companies that quietly dropped out of the thread were the ones running normal product engineering orgs that no longer need to grow.

Two external numbers back this up:

• US tech listings sit about 36% below the February 2020 baseline (Indeed Hiring Lab, July 2025).
• General software engineering postings are down 49% from that baseline, while ML engineer openings are up 59%.

So HN's 2.2:1 is the healthiest slice of a market where general SWE demand has roughly halved. Fed governor Christopher Waller put the macro bluntly: "We're close to zero job growth. That's not a healthy labor market," attributing it to CEOs who are "not hiring because we're waiting to try to figure out what happens with AI." At a Yale CEO gathering in Manhattan, 66% of leaders said they planned to reduce or maintain workforce size in 2026.

The junior surplus is the real story

The HN ratio is 2.2:1 overall, but for entry-level software engineers it is almost certainly 4:1 to 5:1, because the junior supply pool is now 1.65 times the senior pool while HN listings skew senior. In Refolk's index of professional profiles, there are roughly 353,689 US professionals currently titled "Software Engineer" at entry level, against about 213,979 at Senior or Staff.

That ratio only works if juniors get hired at a comparable rate. They are not. Handshake's data shows:

• Student job postings down 15% year over year.
• Applications per job up 30%.
• Class of 2024 submitted 64% more applications per job than Class of 2023.

The mechanism is simple. Companies are posting fewer early-career roles outright, and when they do post one, they often hire a more experienced candidate who dropped a level to get back to work. A 4-YOE engineer taking a mid-level role displaces an intended L3 hire, and the displaced L3 then competes for an L2 posting. The surplus compounds downward.

If you are writing an entry-level resume into this, the correct target is not "look junior and eager." It is "look like the cheapest version of a mid-level hire." Lead with the one shipped project that touched a real user, name the stack, name the metric. That rewrite, posting by posting, is the work Refolk takes off you: paste the job description, get your own resume back rewritten against it, with a fit score that tells you whether the application is worth sending at all.

The 15,118 number: AI as a hard pivot

AI engineering is the only slice where supply is genuinely scarce: only about 15,118 US professionals currently carry "AI Engineer" or "Machine Learning Engineer" as a current job title in Refolk's index, against ML openings up 59% from the 2020 baseline and 500,000+ AI-related roles globally. That is roughly one AI engineer for every 23 entry-level SWEs.

This is not a nudge toward tacking "prompt engineering" onto your skills section. It is the math behind why a hard pivot moves your odds and a soft one does not. If you can honestly show one shipped feature that touched a model (retrieval, evals, fine-tuning, an agent loop in production), the applicable-pool math changes more than any formatting tweak will.

Named entities actually expanding engineering headcount over the two years to mid-2026:

• Ramp: +94%
• Wiz: +84%
• Datadog: +68%
• Rippling: +55%
• Figma: +41%
• Google: 62% more engineering postings in H1 2026 vs H1 2025

And the counter-example: Meta dropped off the top 20 companies by open software engineering positions for the first time since 2018.

Why sending more applications stopped working

When supply doubles, per-application response rate roughly halves, so doubling your volume just restores the old baseline while burning your week. One recent data point making the rounds: a 4-YOE engineer sending 200 applications per month was getting three responses. That is a 1.5% response rate on a volume that leaves no time to tailor anything.

Here is the arithmetic plainly:

1. In April 2026, 1.2 candidates per company. Call baseline response rate R.
2. In September 2026, 2.2 candidates per company. Response rate drops to roughly R/1.83.
3. To hold total responses constant, applications must rise 83%.
4. At 83% more volume, each application gets 55% of your old tailoring time.
5. Less tailoring lowers response rate further, so the real multiplier is worse.

The lever that actually moves the output is targeting and tailoring per application, not count. Tailoring in this market means three concrete things:

• The job's own verbs appear in your bullets, truthfully, in the first half of the resume.
• The exact stack in the posting appears in your skills block, not three stacks that overlap it.
• The cover letter names the product or the team's recent shipped work, not the company's mission statement.

That last one is where candidates lose hours per week. Drafting a cover letter from the posting and your real history is the only version of this task that scales to 20 applications in an evening without the quality collapse that makes 200 applications return three replies.

The concentration risk nobody is pricing

Fewer companies posting more roles per post is not opportunity; it is a pile-on. The same ~240 names are now receiving the applications that used to spread across ~361.

Think about what that does to a hiring funnel at Ramp or Vanta. If the top-of-funnel doubles against a fixed recruiter headcount, screening gets more mechanical, not less. ATS keyword matching gets stricter. The gap between "I could do this job" and "my resume parses as doing this job" widens.

Two tactical implications:

• Stop applying to every role at the same company. Pick the single opening that matches your history best and send one strong application. Three mediocre applications to the same recruiter signal confusion.
• Treat the 240 as a shortlist, not a long list. Build a weekly rotation through named employers who are actually hiring, with a resume variant per company family (infra, dev tools, fintech, AI platforms). Reusing the right variant on the right posting is 80% of the quality lift.

A tactical plan for the 2.2:1 market

The move set that still works has four parts: cut the target list, rewrite per-posting, pivot one axis toward AI if you honestly can, and send fewer, better applications per week.

1. Cut the target list to employers that posted twice this quarter

A company that shows up in two consecutive monthly HN threads is a real hiring signal. One appearance is a trial balloon. Build your list from the companies that posted in both the September and October 2026 threads, then layer in the scaleups growing engineering fastest (Ramp, Wiz, Datadog, Rippling, Figma, Google).

2. Rewrite the resume per posting, not per company

A per-company resume is 2024 thinking. In a 2.2:1 market, the posting-level verbs and stack list are what the ATS and the recruiter screen for. If a posting mentions "distributed systems" and yours says "backend services," you are being filtered out by a word. This is the exact friction Refolk removes: it reads the posting, writes your resume against it from your actual history, drafts the cover letter, and scores how well you fit before you spend an evening on a long-shot application.

3. Make the AI pivot explicit if it is honest

Add a single project line at the top of the experience section that names a model, a dataset, and a shipped outcome. One line. If you cannot write that line truthfully today, spend a weekend building something you can.

4. Cap applications at 15 per week, tailored

Fifteen tailored applications outperform 150 generic ones in this ratio environment. The arithmetic above is why. Use the time you save to prepare for the interview loops you do get, because at 2.2:1 the second-round conversion is where the market gets won.

FAQ

Is the HN "Who is hiring?" ratio representative of the broader market?

No, it is a best case. HN's hiring thread skews toward profitable, AI-adjacent startups that still treat HN as a signal channel. US tech listings overall sit about 36% below the February 2020 baseline, with general software engineering down 49% while ML roles are up 59%. If HN is at 2.2 candidates per role, the segment of the market you cannot see on HN is almost certainly worse.

Does this mean I should stop applying to junior roles?

No, but understand the ratio is likely 4:1 to 5:1 for entry level, not 2.2:1, because the junior pool is 1.65 times the senior pool in Refolk's index while HN listings skew senior. The response is to write your resume to look like the cheapest version of a mid-level hire: name the shipped project, the stack, and the metric in the first third of the page, and skip the coursework section.

Should I pivot to AI engineering?

Only if you can honestly show one shipped feature that touched a model. The supply pool is tiny (around 15,118 US title-holders) against openings up 59% from baseline, so the pivot genuinely moves your odds. A cosmetic pivot (adding "LLMs" to your skills section without a project behind it) does nothing, because the screen for AI roles is specifically evidence of shipped work.

How many applications should I send per week in a 2.2:1 market?

Around 15, each tailored to the specific posting, is a defensible cap. The math of a doubled candidate pool means per-application response rate roughly halves, so volume alone cannot compensate without destroying your tailoring quality. The engineer who sent 200 applications and got three responses is the cautionary tale: at that volume, nothing can be customized, and generic applications are exactly what the market is now filtering out.