The robot housekeeper needs an operator.

On September 17, 2026, Figure announced Helix 2.5 and reported testing three behaviors in 30 unfamiliar Bay Area homes: tidying living rooms, folding towels, and making beds. Figure says no training data was collected in those homes. The tasks themselves were trained elsewhere. Its reported full-task success rate was 56%, with no partial credit. These are Figure’s own evaluation results. Read Figure’s evaluation ↗

That makes the announcement relevant to a service business. A system that needs to learn every customer’s furniture and floor plan separately would carry a substantial setup burden. Transferring a learned task to an unfamiliar room could change how an operator introduces the service to a new household. The remaining failures matter just as much when deciding what to promise.

Figure’s earlier Figure 03 announcement described a robot designed for domestic use, including removable, washable coverings and wireless charging that allows it to dock itself. Those are manufacturer descriptions of the hardware, not evidence that a complete housekeeping service is available. Read the Figure 03 announcement ↗

Still, they point toward useful operational questions. Who prepares a robot for its next visit? Where does it charge? What happens when it stops halfway through a task? A household should be able to ask for help without becoming the maintenance department for the machine in the hallway.

A managed service could, in principle, put that responsibility with an operator. The customer would agree on a defined set of jobs. The operator would handle setup, support, and checking the result. That is a possible business model to investigate, not a service specification Figure has established in these announcements.

For a future pilot, we would start with a narrow routine in a home that has been assessed for it. A defined towel-folding task offers a clearer starting point than an open-ended instruction to clean everything. The customer and operator can agree on what completion looks like and what should be left alone.

That small promise needs an operating plan:

1. Choose the task. Name the objects, rooms, and conditions that are in scope.
2. Define a stop. Make it clear when the robot should pause and ask for help.
3. Assign the response. Give the customer a person or team responsible for an interrupted job.
4. Check completion. Confirm the agreed outcome and record anything still unfinished.
5. Explain the data. Tell the household what is recorded, who can access it, and how long it is retained.

These are proposed requirements for an operator, not capabilities claimed for Figure’s current system. They would shape whether the service feels dependable on an ordinary day, when a room is messy in a way nobody anticipated.

The economic case would need the same discipline. A purchase price alone would tell an operator very little. A useful trial would count preparation, travel if the robot moves between homes, charging, supervision, interrupted tasks, maintenance, and checking the work. Those are items to measure, not costs we can responsibly estimate from a demonstration.

A pilot could reveal that the right model keeps a robot in one household. It could point toward a staffed service with a carefully limited task list. It could also show that support consumes too much time to make the offer worthwhile. Each result would be useful if the trial were designed to answer the business question before a broad promise was made.

The prospect is appealing because the desired outcome is so ordinary: a home that takes less effort to keep in order. A future operator would earn trust by delivering that outcome consistently, explaining the limits clearly, and taking ownership when something needs attention.

For a business considering this space, the next step is a written service brief: the customer, the task, the completion standard, and the evidence needed before a pilot. Figure’s progress makes that exercise worth doing. A credible offer begins with a promise specific enough to test.