Two founders of an embodied AI startup spent a month arguing about their first critical hire after securing angel funding. One wanted an algorithm engineer: without a VLA model, a world model or reinforcement learning, even a well-built machine would remain a remotely controlled toy. The other wanted a hardware engineer: without a body, actuators and sensors, the algorithm would have nothing useful to run on.
They hired for algorithms first. Six months later, the simulation looked good but the real robot did not. Motor delays, sensor noise and inadequate mechanical rigidity prevented the software from behaving as expected. Recruiting a hardware team took another three months, and the overall programme was delayed by six months.
The first appointment can shape the team's difficulties for a long time. The answer depends on what the company needs to prove next.
Start with the uncertainty being tested
An algorithm-led company may be developing VLA models, world models or reinforcement-learning policies on an existing robot platform such as Unitree's H1 or AgiBot's Expedition A2. If it does not intend to build its own complete robot in the near term, an algorithm hire may address the central risk: whether the method works in the physical environment.
A company whose differentiation lies in the robot body, actuators or sensor integration faces a different problem. If the next milestone is a working prototype and a demonstration for investors, hardware capability may be the immediate priority.
The difficulty comes when the founders have not chosen between algorithm research, a proprietary robot and deployment into a customer setting. An algorithm recruit waits for a platform; a hardware recruit waits for requirements. Neither can resolve an undefined company objective alone.
Design algorithms and hardware together
At the 2026 World Robot Conference, Nagesh Puppala, Intel's general manager for physical AI and Robotics, discussed the demands of physical AI. Physical AI, he argued, has to perceive real surroundings and take physical action, making real-time context, deterministic timing, edge power and space constraints, and pre-action safety validation particularly important.
The architecture described puts large-scale training in the cloud, fleet orchestration and analysis at the edge, and immediate perception, control and action decisions on the robot.
For recruitment, the implication is that algorithm and hardware decisions need to meet from the beginning. Developing one in isolation and adapting the other later creates avoidable integration work.
Examples include OFweek reporting on substantial March 2026 funding rounds for companies including MagicLab, Lingchu Intelligence, D-Robotics and Zhangyu Power. The reporting describes their founding teams as having combined algorithm and hardware backgrounds, or having both disciplines represented early.
If the startup can initially add only one person, the founding team must still supply, obtain or explicitly plan for the other capability. The right first recruit is someone who can help complete the immediate validation with the resources actually available.
Product delivery changes what a strong first hire looks like
Figures attributed to Sina's AI reporting put January–May 2026 revenue growth in China's embodied AI industry at 22.4% year on year. The breakdown is 30.1% for robot bodies and complete-machine Manufacturing, 24.5% for AI algorithms and software integration, and 27.9% for systems integration and industry applications.
These figures accompany a shift from technical validation towards commercial deployment. A strong academic record, conference papers, open-source work and laboratory experience remain useful evidence, but they do not necessarily establish the ability to build a product.
The same early role may now call for algorithm work, hardware debugging, simulation-to-real transfer, project delivery and an understanding of the intended customer setting. A purely academic first hire without engineering experience may leave the founders with more gaps to fill later.
Four questions before making the appointment
- What is the central risk now: an algorithm that fails outside simulation, a robot that cannot yet be built, or a deployment that does not work for the customer?
- Is the technical route based on adapting an existing platform or developing a complete robot? The former often puts algorithms first; the latter often puts hardware first.
- Can the new hire complete a useful validation without the other discipline being fully staffed? If not, the milestone or resourcing plan needs to change.
- What does the next commercial milestone require? A physical demonstration and a paper or open-source model may imply different immediate priorities.
These questions are more useful than deciding that one discipline is inherently more important. Both are necessary; the sequence should address the company's current constraint.
UB hunter begins embodied AI team planning by establishing whether the business is validating algorithms, developing a complete robot or implementing an application. That gives the first role a clear puRPOse and shows which complementary skills must already be present.
Before increasing the flow of CVs, put the current stage, technical route, main risk and next milestone on one page. The first critical hire should have a credible route to moving that plan forward.