Harness Research Engineer | 36labs
- Hiring from
- United Kingdom
- Work type
- Hybrid
- Posted
625,827 remote jobs, straight from company career pages
100% free · New jobs every minute
Show job descriptionHide job description
Role: Harness Research Engineer
Company: 36 Labs
Location: London, in the office four days a week (Baker Street now, moving to Primrose Hill within a couple of months). London-based candidates only. No visa sponsorship is available at the moment.
Job Type: Full-time
About 36 Labs
36 Labs is an applied AI research firm in London. It is teaching models what captures human attention on social media, and it starts with creativity. The founders see a creative process as a tree of decisions. LLMs are poor at that, and they are poor at explaining why a creative choice was made.
The product is a harness that sits over any frontier model, especially video generation models. An autonomous marketing service runs on it, from ideation to generation to distribution. The content goes to real audiences, and the results feed a data loop that improves the harness. Clients so far are mainly music companies, including Sony, Warner and Universal. Marketing is a proof of concept. The company does not see itself as a marketing business.
The company is profitable and is closing its first seed round in the next couple of weeks. V3 of the harness is being released internally, and the scope of work is large. The founders are sure of the direction. They need someone to own delivery. Today only one or two people can work on the harness, so you would be one of the first. This is a new role, created by the next stage of growth.
The stack is Python, TypeScript, Postgres, LLM and generative model APIs, and open-weight models.
What you are expected to do:
Continue developing the harness, including the V3 release.
Take ownership of delivering the harness. The direction is set, and getting it built is your job.
Build and run benchmarks and evals.
Own the evaluation pipeline for the harness, across text, video and images.
Turn fuzzy quality questions into validated evaluation models for open-ended agent tasks.
What you need to succeed:
Experience
You have 4+ years of experience and have owned something before, from start to finish.
You are a hands-on engineer in the AI-native era and can work independently.
Agents and evals are the company's main priorities. You have shipped evals for outputs that cannot be tracked automatically, such as open-ended, non-numerical tasks.
Ideally you have shipped an agent system with real users and workloads. It relies on models, calls tools and acts on the results of planning steps. A basic chat interface with retrieval behind it is not enough, and neither is agent work limited to data extraction.
You can work from the London office four days a week and do not need visa sponsorship right now. A visa valid until next June, for example, is fine. A month's notice is fine too. Starting earlier is preferred, though November or December is acceptable.
Nice to have
A degree in computer science, engineering, maths, stats or similar. An MSc or PhD is great. With a BSc only, you need to be really good. The university matters less than what you have built.
Experience with multimodal models or post-training.
A background in applied generative media or creative companies, such as Fal.
Projects beyond university or your day job, and a habit of keeping up with new stacks, models and practices.
How you work
You have high agency. You take the lead and need little hand-holding.
You do things first and ask forgiveness later, and you are open to being disagreed with.
You explain your ideas clearly to the team and leave your ego at the door.
What you get:
Real ownership. You would own the delivery of the harness, which only one or two colleagues can work on today.
Work on a problem few other companies are tackling. The team believes that solving creativity in LLMs would solve the human attention economy, and they see a very high ceiling.
Clients such as Sony, Warner and Universal.
Fast learning at a seed-stage company, which is also a strong line on a CV.
Room to grow into a more senior role. The CTO is expected to step back from the code over the long term.
A pace that is extremely fast and comes without stability. The team sees that as a good thing, and we want you to know it upfront.
How to apply
Send your LinkedIn profile or CV, with a short note on the projects you are proudest of, to your contact at the agency. Use the subject line "Harness Research Engineer, 36 Labs". The process starts with a short pre-screen call with the agency. Next comes a conversation with Benedict, head of operations, then a technical interview of about 30 minutes with the CTO, then a take-home assessment. If the team is torn between strong candidates, there may be a final session on-site in the office.
