· updated
What "Who is hiring?" asks for
Twelve months of Hacker News hiring threads, counted by one script: which AI terms rose, which stayed flat, and what the thread itself did.
Every month Hacker News opens a thread called Ask HN: Who is hiring? Companies post one comment each, a few hundred a month, mostly US and remote software jobs. It is not the job market, but it is the same crowd every month, which makes it a clean place to watch vocabulary move. I count it with a script as part of a wider watch on what AI work employers ask for. The table below is the last twelve threads. It is refreshed monthly; the newest month keeps growing for about three weeks after it opens, so its row is marked partial until the next refresh.
How it is counted
One top-level comment is one post. For each term, a post counts if the term’s pattern matches anywhere in its text, case-insensitive. The share is matches divided by posts in that thread, so months with different sizes compare. The patterns are listed under the table; anyone with the Algolia HN API and a hundred lines of Python gets the same numbers. Posts are cached locally and never republished; only the counts are.
| Month | n | LLM | AI agent | RAG | MCP | Claude Code | Evals | Governance | Azure/AWS AI | Architect | Observability | Agentic coding | Remote | Europe | Contract |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2025-11 | 393 | 14 | 14 | 1 | 0 | 2 | 1 | 0 | 1 | 13 | 4 | 2 | 59 | 15 | 7 |
| 2025-12 | 378 | 11 | 10 | 2 | 0 | 0 | 2 | 0 | 1 | 13 | 4 | 1 | 56 | 12 | 5 |
| 2026-01 | 257 | 14 | 13 | 4 | 0 | 1 | 3 | 0 | 0 | 13 | 4 | 2 | 56 | 13 | 9 |
| 2026-02 | 412 | 12 | 11 | 2 | 1 | 2 | 2 | 0 | 0 | 15 | 5 | 1 | 55 | 15 | 5 |
| 2026-03 | 320 | 10 | 13 | 2 | 0 | 3 | 2 | 1 | 1 | 16 | 5 | 2 | 53 | 17 | 5 |
| 2026-04 | 332 | 15 | 16 | 3 | 1 | 4 | 1 | 0 | 1 | 20 | 5 | 2 | 54 | 14 | 6 |
| 2026-05 | 322 | 13 | 15 | 3 | 2 | 3 | 4 | 1 | 0 | 16 | 6 | 2 | 54 | 13 | 8 |
| 2026-06 | 317 | 14 | 15 | 2 | 1 | 5 | 5 | 1 | 0 | 18 | 6 | 3 | 51 | 11 | 6 |
| 2026-07 | 260 | 14 | 13 | 3 | 2 | 4 | 4 | 0 | 1 | 15 | 3 | 4 | 55 | 15 | 9 |
| 2026-08 | 232 | 16 | 15 | 2 | 2 | 4 | 5 | 0 | 0 | 14 | 6 | 2 | 50 | 14 | 8 |
| 2026-09 | 254 | 15 | 18 | 2 | 4 | 4 | 6 | 1 | 1 | 15 | 6 | 4 | 56 | 17 | 8 |
| 2026-10 (partial) | 261 | 12 | 16 | 3 | 5 | 3 | 5 | 1 | 1 | 14 | 4 | 5 | 56 | 17 | 10 |
Shares in % of posts. n = top-level comments.
Patterns used (regex, case-insensitive)
- LLM
\bLLMs?\b|large language model- AI agent
agentic|\bai agents?\b|multi-agent- RAG
\bRAG\b- MCP
\bMCP\b|model context protocol- Claude Code
claude code- Evals
\bevals?\b|evaluation framework|guardrail- Governance
\bAI Act\b|AI governance|responsible AI- Azure/AWS AI
azure (open)?ai|ai foundry|bedrock|sagemaker|vertex ai- Architect
architect- Observability
observability|langfuse|langsmith|\barize\b|helicone|agent tracing- Agentic coding
agentic coding|\bcoding agents?\b|ai coding (agent|assistant)s?|\bcodex\b- Remote
\bremote\b- Europe
\beurope|\beu\b|\bemea\b|norway|nordic|scandinavia|oslo|stockholm|copenhagen- Contract
\bcontract(or)?\b|freelance|part[- ]time|fractional
Data: Hacker News via the Algolia HN Search API. Last refreshed .
What moved
(this section is rewritten by a person at each refresh; the table is not)
- MCP went from almost nothing to one post in twenty. One post (0.3 %) in November 2025, none in December or January, 4–5 % since September. It is the fastest riser in the table.
- Evals and guardrails more than doubled, 1–3 % through the winter to 4–6 % since June. The posts that name them mostly want someone to make an LLM feature measurable before it ships.
- Agentic coding as a requirement is new. “Coding agents”, “Codex”, “agentic coding”: 1–2 % through May, 3 % in June, then 4, 2, 4 and 5 % in July–October. These are posts that expect you to work with coding agents, not build them.
- AI agents are large and only slightly rising: 10–14 % a year ago, 13–18 % since April. The word is already in most AI posts; the growth is in the words around it.
- RAG is flat at 1–4 %. It did not fall, and it did not follow agents up.
- Governance (AI Act, responsible AI) and named cloud-AI platforms are at 0–1 % on HN. Those asks show up in European consultancy postings, not here.
- The thread itself shrank: July–September had 232–260 posts a month, against 299–331 in the same months of 2025, about a fifth fewer (the 2025 months come from the same script; they are older than the table). Read the shares, not the counts, and read the shrink as a fact about the thread, which may or may not be a fact about hiring.
What this does not say
A match is a word, not a requirement; “architect” catches solutions architects and “remote” catches “no remote”. One company is one post whatever its headcount. The crowd is US startups and remote-friendly firms, so Norwegian or enterprise demand is not in here. And twelve months is one cycle; a line that moved three points may move back. For what the terms look like in practice, the companion note covers the method side: A test that can fail is what “evals” means in a product.