Bloomberg Tech

Jake Loosararian on keeping humans in control of physical AI

Jake Loosararian· CEO at Gecko Robotics
·~6 min·English·Bloomberg
RoboticsAI SafetyAgentsAI Infrastructure
TL;DR

Gecko Robotics CEO Jake Loosararian argues that physical-AI robots earn a place in defense and industry only when a human-set boundary and a watching guard travel with them - the two layers of NVIDIA's new open agent-safety platform, now running on Gecko's Komodo robot.

01Core Architecture

A Box and a Guard

The safety platform does two jobs at once - keep the agent inside the boundaries you set, and post a watcher for the moment it steps outside them.

Part one, make an agent fit within parameters that you set in a box. Right? Part two, a security guard for when that agent misbehaves or acts suspiciously.

— Bloomberg Tech (interviewer)
Key Insight
This is defense-in-depth borrowed from software security - a sandbox that limits what code can do, plus intrusion detection for when something slips the sandbox. Gecko is applying it to an agent that drives a physical robot, where a breakout has consequences in a mine or on a ship, not just on a screen.
02The Data Flywheel

The Moat Is in the Physical World

The durable advantage is not the model everyone can license - it is the datasets you can only collect by running robots through ships, mines, and factories.

deployment into the real world on the physical intelligence side, that's where real large moats are coming from by gathering information. Data sets never exist before to power these abilities to model and simulate what the real world looks like

— Jake Loosararian, Bloomberg Tech
Key Insight
A model is copyable; the field data a robot gathers on the job is not. A rival can license or fine-tune the same foundation model, but it cannot reproduce the deployment records Gecko collects on real contracts - which is why Loosararian puts the moat in deployment, not in the model itself.
03The Real Tradeoff

Safety at Frontier Speed

Loosararian rejects the choice between moving fast and being safe - critical-infrastructure AI has to do both, because the customers are militaries and power grids.

you need to prioritize safety, but you also need to be moving at a pace that's really, bringing to bear the best technology in the world for the people keeping, our borders safe and keeping freedom, for the world.

— Jake Loosararian, Bloomberg Tech
Key Insight
In consumer software, fast and safe trade off cheaply - a bug is a rollback. On a warship or a power grid the same bug is physical, so the two stop being a slider and become hard constraints that both have to hold. That is the real reason a dedicated safety layer, not just a careful model, is the product.
04The Worldview

Used for Good, Not Doom

Against the AI-doom talk from other CEOs, Loosararian frames robotics-plus-AI as a way to pull humans out of danger and keep critical assets running.

certain CEOs about the doom of AI, but I think what's important is this partnership highlights, how, these technologies, robotics and artificial intelligence models, are actually being used for good

— Jake Loosararian, Bloomberg Tech
Key Insight
The optimism has a commercial edge: Gecko sells exactly the deployment-and-safety layer that makes 'used for good' credible to a buyer who fears the downside. The claim is not wrong, but the worldview and the product are the same pitch - safety is what makes the robots sellable into military and energy sites at all.
05On the Ground

Komodo in the Field

The abstract pitch has one concrete case - Gecko's Komodo robot on military contracts, running under the NVIDIA safety platform - with energy, mining, and manufacturing named as the broader critical sectors it could reach.

keep ships out of dry dock and in the fight and ensure that we can, deter conflict around the world.

— Jake Loosararian, Bloomberg Tech
Key Insight
'Keep ships out of dry dock' reframes the robot's value as uptime, not autonomy - the pitch to a defense buyer is not smarter machines but assets that stay in service. That is why the safety layer matters commercially: a robot trusted to run unattended is one that keeps expensive infrastructure available.