1X Technologies NEO Humanoid

Vendor and Product Reality

1X Technologies, formerly Halodi Robotics, is led by Bernt Bornich and operates engineering centers in Moss, Norway and Sunnyvale, California. The company shipped EVE — a wheeled humanoid torso — into industrial security and logistics pilots, and used those deployments to build the manipulation and tele-operation stack that NEO inherits. NEO Beta was unveiled in 2024 with an explicit consumer-home target, soft outer shells, tendon-driven actuators, and a stated design intent of operating safely around humans without the rigid-body force profile of competitors like Figure or Tesla Optimus.

The OpenAI relationship is both financial and architectural. OpenAI's Startup Fund led 1X's Series B, and the companies have publicly described joint work on language-conditioned manipulation and embodied reasoning. NEO is positioned as a platform on which OpenAI-class models drive low-level control through a learned policy rather than through scripted behavior trees, and the consumer roadmap depends on that learned stack generalizing across the long tail of household tasks.

The product story 1X tells publicly is one of in-home assistance: laundry, tidying, fetching, basic eldercare presence. That story implies multiple humanoids per home over time, humanoids cohabiting with non-humanoid robots like vacuums and lawnmowers, and humanoids visiting from service providers. The fleet topology is mixed by construction, and the operator — the human in the home — is non-technical, non-credentialed in any robotics sense, and emotionally invested in outcomes the robot may not understand.

Architectural Gap

NEO's control stack assumes a single tight loop between the household operator's natural-language instruction and the on-board policy that executes it. That loop works for unambiguous tasks in a single-robot home, but it breaks in two directions as the deployment matures. Outward, when a second humanoid or a non-1X robot enters the home, there is no architectural object that represents the operator's intent at a fidelity multiple agents can consume; each robot resolves the instruction independently against its own model and its own world view. Inward, when the operator's intent is partial — "tidy up before my mother arrives" — the policy must guess at fidelity, and it has no graduated channel for asking the operator to refine.

The mixed-fleet case is where the gap shows first commercially. A household running a NEO, a Roomba j7+, a Husqvarna Automower, and a delivery robot from a grocery service has four agents whose work composes — the NEO does not vacuum where the Roomba is currently running, the Automower does not block the delivery robot's arrival path — and today that composition is achieved either through brittle pairwise integrations or not at all. None of the agents can read the operator's intent at a shared fidelity.

The safety case is where the gap shows first regulatorily. Consumer-AI regulation, from the EU AI Act's high-risk categories through emerging US state-level humanoid statutes, is converging on a requirement that operator intent in safety-relevant actions be auditable and credentialed. NEO's current stack records what was instructed and what was done; it does not record what authority graded which fidelity tier and what the operator was given the chance to refine.

What the Operator-Intent Primitive Provides

The operator-intent primitive supplies graduated fidelity tiers for mixed-fleet coordination. Operator intent is modeled as a structured object with explicit fidelity levels — coarse outcome ("home is presentable"), task decomposition ("laundry done, dishes put away, floor clear"), action constraints ("do not enter the nursery before 7am"), and operator-acknowledged risk acceptance ("you may move the espresso machine"). Each tier is signed by the operator at the moment of declaration, and the agents executing against the intent declare which tier they are operating at and what they have escalated for refinement.

Multi-fleet intent fusion is the operational core of the primitive. The same operator intent object is consumed by every agent in the home — NEO, the vacuum, the lawnmower, the visiting delivery robot — and each agent contributes its planned actions back to the object so that conflicts surface architecturally rather than through physical near-misses. The fusion engine resolves authority precedence (the operator's nursery-exclusion overrides any agent's tidying plan), surfaces fidelity gaps (no agent has claimed responsibility for the espresso machine), and produces an attestable record of what the operator authorized and what the fleet did with that authorization.

The graduation matters because consumer operators do not specify intent at uniform fidelity. The primitive lets the operator stay coarse where they want to and forces refinement only where safety, multi-agent conflict, or regulatory audit demands it. The operator's experience is closer to delegating to a household manager than to programming a robot, and the regulatory record is closer to a signed instruction than to a log file.

Composition Pathway

A pragmatic adoption path begins inside the NEO control stack. 1X exposes the on-robot intent object as a first-class artifact rather than as an internal interpretation of a natural-language string. The OpenAI-class model that today renders an instruction into a policy plan instead renders it into a graduated intent object, signs that object on behalf of the operator after acknowledgment, and runs the policy against the object. NEO's behavior does not change for simple instructions; it gains structure for compound and safety-relevant ones.

Multi-NEO homes compose next. As 1X ships second and third units into the same household, the intent object becomes the shared coordination surface, and the units arbitrate through declared fidelity rather than through ad-hoc co-presence logic. This is where the substrate first earns its keep without external partners — 1X owns both endpoints and can demonstrate the architecture cleanly.

Cross-vendor mixed fleet is the third layer. iRobot's Roomba, Husqvarna's Automower, Amazon's Astro, and grocery-delivery robots all subscribe to the same operator-intent object and contribute their planned actions back. The integration contract is the intent schema, not a bilateral API per vendor pair, and 1X — having shipped the substrate first — becomes the natural reference implementation. The household operator interacts with one intent surface; the robots beneath it coordinate without operator involvement.

Commercial Implication

The consumer-humanoid commercial story today is unit sales: 1X sells NEO units at a hardware-plus-subscription price, and revenue scales with units in homes. The operator-intent substrate adds a second surface: 1X licenses the intent platform to peer robot vendors who need to participate in NEO households, and to insurers, regulators, and home-services providers who need to read or write intent on the operator's behalf. That platform revenue is uncorrelated with NEO unit sales and scales with the household graph rather than with the 1X manufacturing line.

Insurance is the most concrete near-term participant. Home insurers who today cannot underwrite humanoid-related liability gain a signed operator-intent record that disambiguates which actions the operator authorized and which the agent escalated. The insurer becomes a co-reader of the intent object and a counterparty to 1X's safety claims, and the resulting policy is materially cheaper for the household. That price difference is itself a sales driver for NEO against humanoid competitors that ship without the substrate.

The defensive case is regulatory. Consumer-AI regulation in the EU and the United States is moving toward affirmative requirements for credentialed, graduated operator authority in safety-relevant household actions, and humanoid platforms will be among the first products tested against those requirements. 1X that ships operator-intent first defines the compliance reference; 1X that does not inherits whatever Figure, Tesla, Apptronik, or a regulator-blessed third party defines.

Licensing Implication

Operator-intent is an Adaptive Query architectural primitive, and the substrate's value depends on the intent object being identical across vendors. A NEO and a Roomba and an Astro that each interpret a coarse household instruction through divergent intent schemas produce no composition; they produce three uncoordinated agents in the same room. A licensed implementation guarantees the schema is shared; parallel reimplementations guarantee it is not. For a consumer-humanoid vendor whose product depends on cohabiting with non-1X robots, the license is the mechanism that makes cohabitation work at all.

Taken early, the license also gives 1X first authorial position over the fidelity tiers, the escalation semantics, and the operator-acknowledgment ceremony that the rest of the consumer-robot industry will eventually conform to. Taken late, the license is still available, but the schema reflects the priorities of whoever moved first. Given 1X's OpenAI alignment and stated consumer-AI safety posture, the architectural opportunity to author that schema is unusually well-matched to the company's existing positioning, and the cost of ceding it to a peer is correspondingly high.