Manicule
AgentRel — Devrel For Agents
Engineering Intern
About the role
Manicule is building AgentRel: systems that help companies become sources AI agents cite and recommend.
We’re solving two problems: understanding how LLMs choose which content and brands to surface, and using AI to produce genuinely useful technical conten, not slop.
I’m looking for an intern who already finds this interesting. Maybe you’ve tested why ChatGPT cites one source over another, built your own agents, or obsessed over making model output less generic. No professional experience is required, but TypeScript knowledge and proof of curiosity are.
What you’ll do
- Ship TypeScript features used by real customers.
- Improve how our agents gather context, draft content, and learn from human review.
- Turn editorial judgment into workflows and quality checks.
- Work directly with founders and customers.
You won’t spend three months on a fake intern project. You’ll ship to production and take on responsibility as quickly as your output earns it.
First 30 days
Probably the highest thing on my list is building an extremely durable and powerful context system.
Our agents need to gather context from dozens of sources, from a company’s Slack and Linear to GitHub and its docs, and use it to produce amazing work on the first try. These sources contain huge amounts of noisy, constantly changing information. You’d build, eval, and improve this production infrastructure.
Bonus points if you’ve worked in context management before.
Details
- Up to three months in person in San Francisco.
- Remote may be possible for the first month.
- Possible full-time offer.
- Meals at the office.
- Unlimited GPT and Claude, including Fable.
About Manicule
Manicule is trying to solve a problem most people have given up on: getting AI to write well.
Everyone can get a model to produce words, but it feels so empty. Almost no one can get it to produce writing with a point of view, a logically sound structure, and a unique tone.
We want to solve these problems at scale. That means turning the judgment of strong writers into something repeatable that an AI can actually run on. Part of the work is engineering: prompts, evals, the harness around them. Part of it is editorial taste. Most of the work is experimentation, because nobody has the answers yet, and the frontier moves every few weeks.
We also think the audience for writing is changing. More and more, what we produce is read by agents as much as by people, and we're already working on catering to them.
We're small, we hire rarely, and people stay. If the idea of teaching a machine to write the way you'd want to read excites you more than it worries you, we should talk.
