Figure AI Helix 02 - Musk Asks if It's Autonomous

On March 9, Figure AI posted a video of its Helix 02 robot tidying a living room without human guidance. Within hours, Elon Musk posted two words on X: "Autonomously or remotely operated?" Brett Adcock, Figure AI's founder and CEO, replied with two of his own: "Fully autonomous."

The exchange captures something real about humanoid robotics in early 2026. The demos are extraordinary. The skepticism is earned. The gap between the two is where most of the story lives.

TL;DR

Figure AI's March 9 living room cleanup video went viral; Musk publicly questioned whether it was teleoperated
Adcock confirmed fully autonomous operation - replacing 109,504 lines of hand-coded C++ with a three-layer neural architecture
Figure has raised roughly $1.75B total at a $39B valuation as of September 2025, backed by Nvidia, Microsoft, Amazon, and Jeff Bezos
Chinese competitors AgiBot and Unitree hold an estimated 58% of current humanoid robot deployments

What Helix 02 Did in That Living Room

The March 9 video extended a sequence that started in January, when Figure published footage of Helix 02 completing 61 sequential tasks in a kitchen over four uninterrupted minutes - walking to a dishwasher, unloading dishes, stacking them in cabinets, loading the machine again, and pressing start. No teleoperation, no scene resets.

The living room demo broadened the manipulation challenge substantially. The robot used a spray bottle to wet surfaces, then wiped with a towel. It lifted a bin while collecting blocks, tucked a container under one arm to free both hands, tossed pillows onto a couch, picked up a TV remote and reoriented it, pressed the power button, and navigated around furniture the whole time.

Adcock described the engineering shift behind this:

"We would have never been able to do a quarter of what you saw today with code heuristics."

That quote carries more weight than it seems. For years, robotics has relied on hand-written control logic - explicit rules telling a robot how to move in specific situations. Helix 02 replaced 109,504 lines of that code with three stacked neural networks.

Helix 02 robot loading a dishwasher autonomously in a kitchen demo The January 27 kitchen demo showed 61 sequential loco-manipulation tasks completed without human intervention over four minutes. Source: figure.ai

How the Three-Layer Architecture Works

System 0 runs at 1,000 times per second. A 10-million-parameter network trained on over 1,000 hours of human motion data across 200,000 parallel simulation environments handles balance, contact management, and whole-body coordination.

System 1 runs at 200 times per second. It takes inputs from head cameras, palm cameras, fingertip tactile sensors, and proprioception data, then outputs joint-level commands for the robot's entire body. Tactile sensors can detect forces as small as 3 grams.

System 2 handles semantic reasoning - understanding the scene and producing task sequences: "walk to the dishwasher," "pick up the red bowl," "place it on the counter."

The four-orders-of-magnitude span between millimeter-scale finger movements and room-scale navigation previously required entirely separate code subsystems. Helix 02 handles both in a single integrated pipeline. Nvidia, which has been pushing into physical AI through its Alpamayo robotics model, backed Figure AI's Series C specifically because of this kind of architecture.

The Money Behind the Robot

Figure AI has raised substantially in its three-year existence. The company's February 2024 Series B pulled in an unusual roster of technology firms - including Microsoft and the OpenAI Startup Fund at a time when those two had interests in common.

Round	Date	Amount	Valuation	Key Investors
Series A	May 2023	$70M	$500M	Parkway Venture Capital
Series B	Feb 2024	$675M	$2.6B	Microsoft, OpenAI Startup Fund, Amazon, Jeff Bezos, Intel Capital
Series C	Sep 2025	$1B+	$39B	Nvidia, Brookfield, Intel Capital, Salesforce, Qualcomm, LG, T-Mobile

Total raised is about $1.75B across all rounds. The jump from a $2.6B valuation in February 2024 to $39B in September 2025 - a 15x increase in eighteen months - reflects the broader market excitement around humanoid robotics more than any disclosed revenue milestone. Figure AI hasn't reported commercial sales figures.

That 15x move wasn't an isolated case. February 2026 broke records for AI venture capital globally, with $189B deployed in a single month. Humanoid robotics valuations have moved in the same direction.

The Musk Question

Musk's skepticism had a specific target. Tesla has been showing Optimus footage since 2022, and that footage has repeatedly attracted questions about teleoperation. In multiple early demos, observers noted movement patterns and reset behaviors that suggested human assistance. Musk has repeatedly described Optimus as Tesla's most valuable long-term asset.

When someone with that track record questions your autonomy claim, the question lands differently than it might from a random account. Musk - through xAI's expanding footprint - is also building AI systems that compete for some of the same talent and capital that funds humanoid robotics. He has incentives beyond neutral curiosity.

Adcock's response was unambiguous. He didn't say the AI is "mostly autonomous with some supervision." He said "Fully autonomous." Figure AI hasn't offered third-party verification - no independent audit, no raw sensor logs, no second camera angle. That doesn't mean Adcock is wrong. It means the industry has a trust deficit that individual assertions alone can't fill.

Counter-Argument

The strongest version of the skepticism doesn't hold up well against Figure's actual documentation.

First, the architecture is specified in detail. Specific parameter counts, inference frequencies, and training data volumes are the kind of claims that invite reproducibility checks. Vague autonomous claims normally come with vague technical descriptions. This one doesn't.

Second, Figure ended its OpenAI partnership in 2025 after deciding the collaboration wasn't moving fast enough. Adcock's comment at the time - "our team just ran circles around them" - suggests a company either truly ahead of the collaboration's output or one willing to take a reputational risk on a false claim. The technical documentation supports the former.

Third, the January kitchen demo was more demanding than the March living room footage. Sixty-one sequential tasks including precise 5ml syringe dispensing and singulating small metal components from clutter are harder to fake than spraying a table and tossing a pillow.

Adcock was careful about timelines even while defending the demo: "Until I feel safe enough to have it there with free reign, it's not ready for everyone." He estimates surgical-level hardware dexterity will be ready by end of 2026, but the AI systems require substantially more training data. That's not the language of a company trying to oversell.

What the Market Is Missing

The Musk exchange dominated the news cycle. The competitive landscape mostly didn't.

As of early 2026, AgiBot holds approximately 31% of humanoid robot deployments, and Unitree holds another 27%. Tesla Optimus sits at around 5%. Figure AI doesn't appear in current deployment share figures because it hasn't shipped commercial units at scale. The product heading toward market - Figure 03, powered by Helix 02 - has no announced release date.

Goldman Sachs projects the humanoid robot market at $38B by 2035. Morgan Stanley's scenario extends to $5 trillion by 2050. These aren't production forecasts. They're analyses built on adoption assumptions nobody has verified against actual purchase behavior.

The question in March 2026 isn't whether Helix 02 is autonomous. Adcock says it is, and the documentation supports that. The real question is whether a $39B valuation buys Figure enough runway to close the commercial gap before AgiBot and Unitree flood the market with units that are adequate at lower prices. The viral video settles one argument. It doesn't settle that one.

Sources: