We are assembling a founding team to lead the next era of AI: digital humanoid workers that deliver real business outcomes.
Tech companies are software and operations. Software scales. Too much of the work behind it is still manual. We are building digital humanoid agents to automate operations at scale.
Founded a marketplace, built its software, and ran its operations. We lived the daily reality of manual work.
Experience inside global enterprise services and operations.
Experience building within enterprise technology organizations.
Prior experience, not organizational affiliation or endorsement.
Individual backgrounds, not organizational affiliations or endorsements.
Our ambition is to set the standard that other AI labs follow. The measure is useful work: digital humanoids that see, reason, act, and carry a mission through to a verified outcome.
We call the distance between frontier intelligence and dependable, affordable execution the Pioneer Lab Gap. Closing it takes invention across the stack: kernels, models, perception, agents, and product. One worker. Coordinated teams. Eventually, autonomous organizations.
We are building toward billions of autonomous digital workers, provisioned on demand for organizations. Reliable, secure, 24/7 work at that scale demands exceptional people in AI, agents, engineering, research, and security. That is the team we are assembling.
Our engineering cost target. Making capable workers widely accessible means pursuing reliability and radical efficiency together. This is a research ambition, not a current price or achieved benchmark.
We are looking for people who have pushed these fields forward, from leading AI labs and research groups to teams that ship at global scale. Come own a hard part of the worker, from first principles to production.
Anthropic / OpenAI / Google DeepMind / Adept / UI-TARS / Qwen / AutoGLM
Computer use made interfaces computable. We take that intelligence across screens and devices: shared context, concurrent actions, and reliable recovery across interfaces.
Browser Use / Skyvern / Browserbase / Steel / Manus / H Company / Microsoft
Multi-agent orchestration, persistent sessions, and recovery across thousands of runs on local and remote devices.
OpenAI / Anthropic / Google DeepMind / Meta FAIR / Mistral / Thinking Machines / Cohere / SSI / xAI
Post-training, distillation, and evaluations with measurable gains in capability, reliability, and cost.
Google / Meta / Apple / Microsoft / Amazon / NVIDIA / Stripe / Palantir / Waymo / Tesla / SpaceX
Products and infrastructure shipped at scale. The performance numbers, failure modes, and decisions behind them.
Carnegie Mellon / Berkeley / Stanford / UW / HKU / Tsinghua / Peking / Toronto / Waterloo / IISc
Original research enabling digital humanoids to use computers, phones, tablets, watches, and browsers concurrently. Parallel and remote device use, shared context, and reliable handoffs across devices.
MIT / Stanford / Harvard / Caltech / Princeton / Carnegie Mellon / Berkeley / Cambridge / Oxford / ETH / EPFL
Deep work in learning, perception, systems, and optimization, paired with the ability to turn an idea into a working system.
These are recruiting interests, not current team affiliations or partnerships.
Exceptional work earns the conversation. Show us the model you trained, the system you made reliable, the benchmark you advanced, or the product people depend on. If your path looks different, the same bar applies. We want to see your work. Include a direct link to a project, demo, paper, or benchmark we can review. Source code is optional.
Explore something you have built, work through a problem with us, and decide what we could build together.
Three stages.
One paid working session.
Many are called.
Few are chosen.
Here’s how we get to know your work—and how you get to know ours.
Explore something you built, improved, or discovered with someone who understands your discipline. Walk us through your contribution, the difficult decisions, and the evidence that it worked. We will also discuss the role and compensation expectations.
Depth & ownershipWork through a prepared problem from your discipline with a future teammate. We provide the environment, tools, and compute. Use AI and documentation, test an idea, and explain what the results tell you.
Judgment & verificationMeet a founder and discuss what you could own, how we work, and the questions still ahead of us. There is dedicated time for your questions about the people, the research, and the company. Decide what we could build together.
A decision on both sidesDuring the working session, we introduce a new piece of information. We want to see how you test your assumptions, update your approach, and explain what you now know.
Bring material you are authorized to share. Public projects, experiments, papers, or a fresh technical walkthrough are welcome. Required interview work fits within the scheduled sessions.
We share the format and evaluation criteria beforehand, and supply the tools and compute. AI and documentation are welcome; understanding and checking the output matters.
We agree the working-session fee and payment timing before scheduling. The default is 90 minutes. If a role calls for a longer session, we discuss it as an alternative and agree the duration and fee upfront.
Tell us what works around your existing commitments, including any access or accommodation needs. Depending on the exercise and team availability, we can discuss remote participation, split sessions, or evening and weekend options.
Use your own device or one we provide. Keep employer accounts, equipment, confidential code, and private data out of the process. If an exercise raises a confidentiality or employment concern, tell us so we can agree an alternative. We will not contact your current employer without your permission.
Exercises are prepared for evaluation; we do not use candidates’ submissions in production. Depending on the role, progress might be a tested change, a well-designed experiment, or a diagnosis supported by evidence. Examples include:
Technical depth, sound judgment, careful verification, learning, and collaboration. Candidates for the same role are assessed against the same criteria. Interviewers record their evidence independently before discussing a decision.
Our target is to complete interviews within five business days of the first conversation, around your availability, and communicate a decision within two business days of the final conversation. If that timing changes, we will let you know.
Seven technical disciplines. One connected research and engineering agenda: expert digital humanoid workers that people can depend on.
A different kind of contribution?