Harness Engineer

VideoClaw is a San Francisco startup building an AI-native video-creation tool: a Mac app where you type a prompt (it works with Claude or ChatGPT) and get a finished video back, using your own GPU to keep it fast and cheap. Founded by Khailee Ng, a Malaysian-born creator-turned-operator who built his country's largest news company and a Groupon clone (acquired in about 4 months), then spent 13 years scaling 500 Global from about $29M to a $2.4B venture platform (still an owner and board member). VideoClaw is his return to building. It is backed quietly by a handful of marquee names he will only disclose to a hire at the very end.

The core team is three: Khailee (vision, fundraising, product and culture), Joseph Chin (primary product and tech decisions), and Sam Yee (COO and ops, hiring, field marketing). The thesis: drop the cost of creating video low enough and far more people create; video is the world's most widely understood language but few can speak it, and VideoClaw lets everyone speak it; win on simplicity and cost, not $200-in-3-days generation. The wedge is content engineering. The first customers are developers and technical people (the 100M+ people vibe-coding on Replit, Lovable, and Bolt who need demo, trailer, and marketing videos), plus GitHub-project and SaaS-testimonial video. The philosophy: every video you want already exists on the web or your phone. Pull those in and edit locally, and only generate the roughly 1 in 10 shots you are missing. The Mac app launched about a week ago with strong early retention. Seed stage, about 1 to 10 people, founded in 2026.

This is the founding engineer who owns and builds the core product while Joseph pushes the leading and experimental edge. The two cover for each other. The must is harness engineering. This is a content engineer and harness engineer seat, not a traditional software-dev one. A normal engineer sees a bug and fixes it. This person designs a system that improves itself: every user feedback click logs a screenshot and the full error context, and a standing agent takes the nightly error logs, fixes things itself, and reports back what it improved for review. You stay plugged into the frontier of models (niche and Hugging Face models that are good at a specific job) and integrate them so it feels seamless. You eventually work with the ML hire to train small on-device models for parts of the creative pipeline, such as a brief-taking agent or a QA agent.

You are a T-shaped generalist, solid on the front end and the back end, whose deep spike is harness engineering. You hold the whole system in your head. This is an IC seat, not a tech-lead or manager seat. The next six months are four levers on the core product: bring down the time a video takes (currently about 15 minutes to 13 hours, using faster models and quick in-between drafts), increase the magic (integrate the best niche models), bring down cost per video (voice-clone and lip-sync renders at about 1/10 the cost, cache repeated web searches), and build the self-improving harness for stability. What matters most is what you have built: a core engineer on a comparable product, or a core contributor to a relevant open-source project.

Bar: harness engineering is non-negotiable. T-shaped, well-rounded, not hyper-specialized. Deeply plugged into niche and open-source models, model integration, and agentic pipelines. You have gone through the fire of building something similar. Company and school are irrelevant. Slope and craft over seniority. A creative side (an instrument, making content, real artistic intuition) is a tiebreaker, not a requirement. Onsite in San Francisco, full-time in the office, including weekends. Visa sponsorship is available. Equity is about 0.1% to 0.7%.