3 startups hiring engineers ask for 4+ years of LLM experience. ChatGPT is 3 years old.
August 2026
Job posts for engineers sometimes ask for several years of experience with large language models, a technology most people first used when ChatGPT launched in November 2022. We wanted to know how common this is, so we checked every live remote job posted by startups and scaleups on Matcha over the last six months, 12,811 posts, for any requirement that attaches a number of years to LLMs, generative AI, AI agents, prompt engineering or RAG.
The short version
34 posts do this, 0.3% of the total, from 27 companies.
What those companies ask for:
- One year: 9 companies
- Two years: 13 companies
- Three years: 3 companies
- Four years: 2 companies
- Eight years: 2 companies
The most common wording is "5+ years of software engineering, including 2+ years working with LLMs in production." Two years takes you back to mid 2024, by which point most teams had shipped at least one feature on top of an OpenAI or Anthropic API.
ChatGPT launched on November 30, 2022, so it is three years and nine months old at the time of writing. Three companies ask for more LLM experience than that:
- Allocate, Machine Learning Engineer: "4+ years of applied ML / LLM engineering in production"
- AZX, Software Engineer: "4+ years of experience building production LLM/agent pipelines"
- Pencil, Head of Agent Architecture: "8+ Years of Direct GenAI Experience, with 2+ Years Leading People or a Function"
No company in our data asks for five, six or seven years.
The two eight-year requirements
Federato, an insurance software company, wants a Staff Machine Learning Engineer with "at least 8 years" as an ML engineer, "with a strong focus on leveraging LLM models over the last 2 years." That is eight years of machine learning and two of LLMs. Machine learning has been an industry job for well over eight years, so this is a normal requirement.
Pencil, an AI advertising company, asks for eight years of direct generative AI experience. Eight years before the post went up is August 2018. For reference:
- June 2017: the Transformer paper is published
- June 2018: OpenAI releases GPT-1
- February 2019: GPT-2
- June 2020: the GPT-3 API opens to a private beta
- November 2022: ChatGPT
- March 2023: Claude
Direct generative AI experience in 2018 means working at OpenAI, Google Brain, or a university NLP lab. Pencil's other open post, for an Agent Architect, asks for three years.
What most posts say instead
Most AI roles in our data do not attach a number of years to LLM experience. Of the 4,304 remote software engineering posts we have, 1,058 mention LLMs, and 34 turn it into a years requirement. Several say the opposite:
- Fieldguide, hiring AI engineers at three levels: "We care more about capability and trajectory than years on a resume."
- Windmill, AI Engineer: "0-5 years of experience" and "we care more about what you've built than years on a resume."
- RxSense, Principal AI Ops Engineer: "We care more about depth and drive than years on a resume."
- Ruby Labs, Senior AI Engineer: "Less experience may be considered for exceptional candidates."
What this means if you are applying
The usual requirement for LLM work at remote startups is two years of shipping features on top of models. One production feature with retrieval, tool calling or an agent loop covers it.
When a post asks for four or more years of LLM experience, the number is above what almost anyone can have. Apply anyway and describe what you shipped.
How we counted
Live remote posts on Matcha from fully remote startups and scaleups, March through August 2026, excluding posts marked expired or quarantined. We searched the original posting text for "N years" within a short distance of LLM, large language model, generative AI, GenAI, agentic, AI agents, prompt engineering, foundation models or RAG, then read every match. We removed founder bios, product marketing copy ("15 years of data expertise with agentic AI"), general experience requirements that happened to sit near the word "agent", and one law firm asking for an LLM, which in that context is a Master of Laws. Counts are per company, so a company that reposts the same role in ten countries counts once. ChatGPT's age is measured from its public launch. The GPT-3 API had a private beta from June 2020, so a four-year claim is possible for a small number of early users.