Q&A with Skild AI CEO Deepak Pathak on building a general-purpose brain for robots, standing out among big tech's robotics efforts, the path to AGI, and more
Deepak Pathak has spent 15 years trying to solve one of the hardest problems in AI: getting machines to move through and manipulate the physical world.
Context & Ripple Effects
Skild AI’s general-purpose robotics thesis was already backed by a $300 million Series A for its foundational robotics model, placing it among companies attempting to build reusable intelligence rather than a robot for one task.
The interview joins an ongoing debate over whether robotics progress needs more than model scale: Google DeepMind’s robotics leadership has also discussed general-purpose robots, while AGI coverage has emphasized continual learning as a prospective breakthrough.
First-order effects
- The Q&A gives Skild AI a clearer public position: it is competing on a general-purpose “brain” for physical-world manipulation, rather than on a single robot form factor.
- Deepak Pathak becomes the visible spokesperson for Skild AI’s differentiation against larger technology companies’ robotics programs and for its link between robotics and AGI.
Second-order effects
- The framing raises the comparison point for rival robotics developers: they must distinguish whether their advantage lies in generalizable software, hardware integration, or access to deployment data.
- For prospective customers and hardware partners, the value proposition shifts toward a reusable intelligence layer that could span robot types—though its practical value depends on demonstrated transfer across physical tasks.
Third-order effects
- If general-purpose robotics models prove transferable, value could concentrate in the software, data, and integration layers above commodity robot hardware, while deployment partners become important sources of real-world learning data.
- The story is another test of whether embodied learning is a necessary complement to language-model scaling on the path to broader AI capabilities; the corpus does not establish that outcome.
The trend: Robotics is moving toward foundation-model-style software layers that aim to generalize across machines and physical tasks, while competing with vertically integrated big-tech efforts.