<p>Eighteen months after the first commercial humanoid robot deployments in real workplaces, enough operational data has accumulated for an honest industry assessment. The verdict is more nuanced than the "robots will replace all workers" or "this is expensive theater" camps both predicted.</p>
<h2>Where Humanoids Are Working</h2>
<p>Agility Robotics' Digit units at Amazon fulfillment centers have proven most consistently successful at one specific task: moving totes from conveyor lines to shelving. The task requires bipedal mobility to navigate human-designed aisles, basic manipulation to handle standardized totes, and reliable repetition over long shifts. Digit's performance on this task has improved significantly from initial deployment — uptime is above 95% for tote-moving, and cost per operation is approaching the economics of human labor in high-wage US markets.</p>
<p>BMW's deployment of Figure 01 for sheet metal transfer — moving metal blanks between stamping stations — is operating in structured manufacturing environments where the task is highly repetitive and the workspace is engineered for robot accessibility. These deployments are working.</p>
<h2>Where Humanoids Are Struggling</h2>
<p>Dexterous manipulation — tasks requiring fine motor control, variable grasping, or handling of diverse item types — remains a significant weakness. Picking individual items from unstructured bins (the Amazon picking problem) is still slower and less reliable than humans. General-purpose factory work that requires switching between varied tasks every few minutes is not working commercially at scale.</p>
<h2>The 5-10 Year View</h2>
<p>The realistic path: humanoids will expand from 2-3 specific, structured task types to perhaps 8-12 by 2030, as manipulation and mobility improve and as workplace environments are incrementally adapted to reduce variability. Full general-purpose factory labor replacement remains a 10-15 year horizon at minimum, and likely longer given the long tail of task variability in real workplaces.</p>