What counts as AI-related?

A post is AI-related when it explicitly mentions AI, ML, LLMs, agents, RAG, embeddings, or known AI infrastructure terms.

What is not proven by this report?

It does not prove total market demand. It measures visible startup hiring language on HN.

AI Hiring Reality Report

AI became normal hiring language.

In the first half of 2026, 58.5% of analyzed Hacker News hiring posts mentioned AI-related work. In the same months of 2022, the share was 18%.

That is a sharp change in how startups describe their work. It is not proof that six in ten openings are AI jobs. One company post can advertise several roles, and an AI mention can describe a product, a tool, or the company’s market.

The long view

The long rise, then the break

AI language grew slowly through the 2010s, sat near 18% from 2018 through 2020, then accelerated after 2022. By 2025, 49.8% of analyzed posts mentioned AI-related work.

Before the surge

The signal was already there

AI-related language was not born with the current boom. It appeared in 5.3% of analyzed posts in 2011 and reached 18% in the first half of 2022. The earlier signal was smaller, but persistent.

The break

The pace changed sharply

The annual share rose to 36.2% in 2024 and 49.8% in 2025. The 2025 increase alone was 13.6 percentage points.

The scale

3.2 times the 2022 share

Comparing the same six-month window, the 2026 share is 3.25 times the 2022 share. The increase is 40.5 percentage points.

Recent movement

The latest signal

The most recent complete month, June 2026, recorded 176 AI-related posts out of 287 analyzed posts. That is 61.3%.

Evidence

What this does and does not prove

The numerator is the number of analyzed top-level hiring posts with explicit AI-related language. The denominator is every analyzed hiring post in the same period. A post counts once, even when it advertises several openings.

Annual AI-related hiring language with post counts and shares
Period	Analyzed posts	AI-related posts	Share
2026 H1	1,831	1,071	58.5%
2025	3,786	1,885	49.8%
2024	3,692	1,337	36.2%
2023	3,888	1,040	26.7%
2022	7,201	1,369	19%
2021	9,780	1,588	16.2%
2020	6,750	1,241	18.4%
2019	8,000	1,456	18.2%
2018	9,160	1,689	18.4%
2017	8,921	1,477	16.6%
2016	7,406	943	12.7%
2015	6,550	652	10%
2014	3,892	403	10.4%
2013	3,015	272	9%
2012	2,266	193	8.5%
2011	1,586	84	5.3%

View the latest 24 complete months

Latest 24 complete months of AI-related hiring language
Month	Analyzed	AI-related	Share	95% interval
2026-06	287	176	61.3%	55.6% to 66.8%
2026-05	298	163	54.7%	49% to 60.3%
2026-04	316	190	60.1%	54.6% to 65.4%
2026-03	303	176	58.1%	52.5% to 63.5%
2026-02	386	228	59.1%	54.1% to 63.9%
2026-01	241	138	57.3%	50.9% to 63.3%
2025-12	355	196	55.2%	50% to 60.3%
2025-11	372	217	58.3%	53.3% to 63.2%
2025-10	324	177	54.6%	49.2% to 60%
2025-09	282	145	51.4%	45.6% to 57.2%
2025-08	286	156	54.5%	48.8% to 60.2%
2025-07	313	156	49.8%	44.3% to 55.3%
2025-06	337	166	49.3%	44% to 54.6%
2025-05	278	129	46.4%	40.6% to 52.3%
2025-04	283	147	51.9%	46.1% to 57.7%
2025-03	325	138	42.5%	37.2% to 47.9%
2025-02	369	147	39.8%	35% to 44.9%
2025-01	262	111	42.4%	36.5% to 48.4%
2024-12	306	129	42.2%	36.8% to 47.8%
2024-11	308	131	42.5%	37.1% to 48.1%
2024-10	312	103	33%	28% to 38.4%
2024-09	287	106	36.9%	31.6% to 42.7%
2024-08	302	118	39.1%	33.7% to 44.7%
2024-07	356	127	35.7%	30.9% to 40.8%

Method

Source: public top-level posts in Hacker News “Who is Hiring” threads. We include posts that name an organization and an opening. AI-related means the post explicitly mentions AI, machine learning, large language models, agents, RAG, embeddings, or known AI infrastructure terms.

Shares use post counts, not opening counts. The intervals show binomial sampling precision around each observed share. They do not include uncertainty from categorization or support claims about the wider hiring market. Hacker News startup hiring posts are a distinct sample, so these findings should not be read as totals for all employers or all software jobs.

The regular monthly series begins in 2011. One earlier observation from November 2008 remains in the download but is omitted from the annual chart because it does not represent a full year.