How to hire AI engineers from what they built

Search for "hire AI developers" and most of what comes back is staffing firms and talent marketplaces. They send you vetted contractors, which works when you need someone fast and for a fixed time. If you want to hire an AI engineer onto your own team, there is a more direct source: the code AI engineers write in public, and the projects they build.

This guide shows where that work lives, who builds the most-used projects (as numbers, from a real search on 27 October 2026), and how to go from a project to a shortlist.

Where AI engineers show their work

Most public AI engineering falls into three kinds of project:

• Agent frameworks: orchestration, tools, memory and multi-agent workflows. LangChain, LangGraph, CrewAI and AutoGen are the best known.
• MCP servers and SDKs: the Model Context Protocol connects agents to tools and data. People who build MCP servers know how agents call tools, and what breaks when they do.
• Inference and serving: running models fast and cheaply. vLLM, SGLang and llama.cpp are the big ones; the work is batching, caching, kernels and quantisation.

Pick the kind closest to what your team will build. An engineer who wrote an agent framework's checkpointing is a different hire from one who tuned an inference engine's attention kernels.

Who builds the popular projects

We ran three searches through AyHire's MCP server on 27 October 2026, each for repos with 1,000 stars or more, pushed since 1 September 2026. For each, AyHire read the 8 most-starred repos and returned the five people who wrote the most of each, with their profile, rank and commit count. Below are the projects, by the commits of their top five people and the share of the one who wrote the most. The links open each repo's contributors page.

Agent frameworks (Python, 70 repos matched)

• langchain-ai/langchain, 148,555 stars: 5,540 commits, top person 25%.
• TauricResearch/TradingAgents, 117,151 stars: 491 commits, top person 92%.
• microsoft/autogen, 61,588 stars: 1,562 commits, top person 30%.
• crewAIInc/crewAI, 60,174 stars: 1,854 commits, top person 40%.
• HKUDS/nanobot, 49,498 stars, created February 2026: 3,632 commits, top person 59%.
• AstrBotDevs/AstrBot, 44,155 stars: 3,959 commits, top person 89%.
• langchain-ai/langgraph, 43,992 stars: 3,019 commits, top person 26%.
• openai/openai-agents-python, 30,334 stars: 1,323 commits, top person 61%.

MCP servers and SDKs (TypeScript, 66 repos matched)

• microsoft/playwright-mcp, 38,708 stars: 491 commits, top person 49%.
• DayuanJiang/next-ai-draw-io, 36,444 stars: 668 commits, top person 90%.
• GLips/Figma-Context-MCP, 16,078 stars: 206 commits, top person 94%.
• modelcontextprotocol/typescript-sdk, 13,661 stars: 988 commits, top person 37%.
• modelcontextprotocol/inspector, 11,164 stars: 3,709 commits, top person 75%.
• mcp-use/mcp-use, 10,816 stars: 1,267 commits, top person 39%.

Two more of the eight are left out here, because an automation account sat among their top five committers. Check every list for accounts like that before you read it.

Inference and serving (any language, 84 repos matched)

• huggingface/transformers, 168,244 stars: 6,113 commits, top person 26%.
• ggml-org/llama.cpp, 132,546 stars: 3,855 commits, top person 53%.
• vllm-project/vllm, 95,696 stars: 3,587 commits, top person 26%.
• lyogavin/airllm, 40,958 stars: 406 commits, top person 97%.
• sgl-project/sglang, 38,274 stars: 5,099 commits, top person 25%.
• jundot/omlx, 23,103 stars, created February 2026: 2,397 commits, top person 91%.
• kvcache-ai/ktransformers, 19,641 stars: 711 commits, top person 34%.

The commit counts are GitHub's (commits on each repo's main branch). We list projects, not people: the developers behind them didn't ask to be on a page about hiring, and AyHire shows what it holds about a person only to signed-in users. Run the search yourself and you get the names, profiles and links: agent frameworks in Python, MCP servers in TypeScript or LLM inference. Each link opens AyHire with the role typed in, on a 14-day trial with no card.

What the numbers say

• Stars don't tell you how many people built something. airllm, Figma-Context-MCP, omlx and AstrBot are mostly one person's work. vLLM, SGLang, LangChain and transformers are spread across a team, with no one above about a quarter of the top five's commits.
• The two kinds tell you different things. A popular project that is mostly one person's shows they can ship and own a product end to end. A top-five place in a team project shows sustained work that passed other people's review.
• New projects get big fast. nanobot and omlx were both created in February 2026. Search with a recent pushed date, and add a created date to find what went viral this year.
• The top of a big project is the hardest to hire. Look further down the contributors page too. Someone ranked tenth in a big project can still own one area of it.

From a project to a shortlist

1. Pick three to five projects close to your work, not the most famous ones. If you are building agents on top of MCP tools, the MCP list is a better start than inference engines.
2. Open each contributors page and note the people whose commits are recent, not only the largest totals.
3. Read what they got merged. In a repo's pull requests, is:pr is:merged author:USERNAME shows their merged work and the code areas it touched. Reviews they gave say as much as code they wrote.
4. Write down what the role needs as must-haves and nice-to-haves, and judge each person against those, not against how famous the project is. Missing evidence is unknown, not a no: most AI work in companies is private.
5. Write one message per person, naming the work you read. Never bulk-email people you found on GitHub: GitHub's Acceptable Use Policies forbid using information from GitHub for unsolicited bulk email.

Our guide to finding engineers on GitHub covers each step in more detail, and reading AI fluency from GitHub covers how to tell whether someone checks what their coding agent wrote.

Doing it with AyHire

AyHire does the reading for you. Describe the role in a sentence, such as "a senior engineer who has built agent frameworks in Python". AyHire plans the GitHub searches and shows them before they run, finds the people who wrote most of the most-starred matching repos, scores the five who wrote the most, and judges them against the role's criteria from their commits, pull requests and READMEs, with a link behind every judgement. Then it tells you who to talk to first, why, and what to check, and who is open to work.

It never contacts anyone for you, and it never uses where someone lives.

AyHire is $99 a month for one person, or Team at $79 a person a month, with a 14-day free trial and no card. See plans and start a trial. It also works from Claude Code, Codex and Cursor over MCP: see Getting started.

Staffing firm or direct hire?

• A staffing firm or marketplace fits when you need someone this month, for a fixed project, and are happy to pay a margin on their rate for the vetting.
• Hiring directly from public work fits when you want someone on your team for years, and want to see what they have built before the first call.

If you compare tools for the direct route, see GitHub recruiting tools, SeekOut alternatives and Juicebox alternatives. If you already pay for LinkedIn, what LinkedIn Recruiter costs puts the two side by side.

Questions people ask

Where can I find AI engineers to hire?

In the projects they build. Search GitHub for agent frameworks, MCP servers or inference engines close to your work, open each repo's contributors page, and read what the people there got merged. Conferences and papers help for research roles; for engineering roles, the code is the best evidence.

What should an AI engineer's GitHub show?

Merged work in projects other people use, recent activity, and clear writing in pull requests and READMEs. For agent work, look for evaluation and error handling, not only demos. A quiet profile tells you little either way, because most company AI work is private.

How do I hire AI agent developers?

Start from agent frameworks and MCP servers, as in the lists above. Judge people on the parts of agents your product needs (tool calling, memory, evaluation, deployment), and ask about the private work their public profile can't show.

Do I need a staffing firm to hire AI developers?

No. A firm saves you the search and the vetting, at a margin. If you can read public work, or use a tool that reads it for you, you can find the same kind of engineer directly.

Sources: AyHire's find_people_behind_repos, three searches run on 27 October 2026 (agent framework in Python, MCP servers in TypeScript, LLM inference), 8 repos each and the top five people of each repo. Stars and commit counts are GitHub's, as of that day. GitHub Acceptable Use Policies.