The AI landscape doesn't move in one direction — it lurches. Some techniques leap from experiment to table stakes in a single quarter; others stall against regulatory walls, technical ceilings, or organisational inertia that no amount of hype can dislodge. Knowing which is which is the hard part. The State of Play cuts through the noise with a rigorously maintained index of AI techniques across every major business domain — classified by maturity, evidenced by real-world adoption, and updated daily so you always know where you stand relative to the field. Stop guessing. Start knowing.
A daily newsletter distilling the past two weeks of movement in a domain or two — delivered to your inbox while the index updates in the background.
Each dot marks the weighted maturity of practices within a domain — hover for a brief summary, click for more detail
AI controlling physical systems across manufacturing, agriculture, construction, healthcare, and service. The largest domain at 27 practices, with industrial robotics and pick-and-place at good-practice. Most practices cluster at leading-edge — sim-to-real transfer, humanoid robotics, and surgical automation are progressing rapidly. Nearly half the practices are advancing, making this one of the most dynamic domains by momentum.
The headline: Robots that do one thing well are delivering real returns in factories, mines, and warehouses. Robots designed to do everything remain overwhelmingly pre-commercial despite $18 billion in investment.
Most large organizations now have some form of physical AI in production -- warehouse robots, autonomous floor scrubbers, quality inspection cameras, predictive maintenance sensors. These narrow-task systems deliver documented 20-50% cost reductions and pay for themselves within two years. A smaller group -- primarily in mining, automotive manufacturing, and large-scale logistics -- is pulling ahead with fleet-scale deployments of hundreds or thousands of robots coordinated by AI orchestration software. The rest of the market faces a closing window: Chinese manufacturers now hold 58% unit share in commercial cleaning robots at roughly half the price of Western competitors, and Asia-Pacific captures 70% of new collaborative robot installations. General-purpose humanoid robots -- the category attracting the most attention and capital -- have yet to produce a single profitable deployment outside tightly controlled demonstrations.
A logistics operator published the first honest humanoid-versus-human comparison. Across five U.S. facilities with 1,200-plus workers, humanoid robots won on repetitive material movement and structured picking but lost on 80% of warehouse tasks including kitting, exception handling, and variable product work. For operations leaders considering humanoid pilots, this maps exactly which tasks justify the investment and which do not.
Stanford confirmed humanoids fail 88% of household tasks. A peer-reviewed assessment found battery life under 8 hours, inconsistent hand dexterity, and AI edge-case failures as binding constraints -- directly contradicting vendor demonstrations of general-purpose capability. This matters because several vendors are pricing consumer humanoid leases at $400-600 per month; the evidence suggests most homes would get minimal productive value at that price.
Tesla admitted Optimus Gen 3 production delays. During Q1 2026 earnings, the company disclosed a four-month factory conversion, 10,000 unique components, and no prior mass-production experience. It declined to provide 2026 production targets and stated only "simple skills" would be attempted initially -- a material downgrade from earlier ambitions.
Japan Airlines began a three-year humanoid pilot at Haneda Airport. The JAL/GMO/Unitree partnership is the first humanoid deployment in a safety-critical regulated environment that requires no facility redesign. It uses simulation-first integration and incremental task rollout, providing a template for infrastructure-compatible deployment in aviation and similar sectors.
Autonomous construction hit cost parity. Sika Brasil 3D-printed a full concrete house in 60 hours at 30% lower cost than conventional building. Heidelberg Materials scaled autonomous haulage to 30 trucks across 6 global sites. Together, these signal that the economics of construction automation are crossing the threshold where pilot-stage caution becomes competitive risk.
Chinese humanoid standards may become the global default. China's MIIT published the first national humanoid robot standard covering safety, behavior, and ethics -- with no Western equivalent at similar specificity. If international bodies adopt this framework, organizations planning humanoid deployments will need to design for Chinese compliance requirements. Track the ISO/IEC response over the next 12 months.
Orchestration software is becoming a standalone product category. LG CNS launched PhysicalWorks for multi-vendor robot fleet coordination; MODEX 2026 surveys show integration complexity has overtaken cost as the top adoption barrier. If you already operate robots from multiple vendors, evaluate orchestration platforms now -- waiting means building custom integration that depreciates as commercial platforms mature.
Sidewalk delivery regulation will tighten further in the U.S. Chicago's ban, Glendale's moratorium, Philadelphia's incident-reporting gap, and Congressional attention all point toward federal oversight of commercial delivery robots within 18 months. Organizations with sidewalk delivery partnerships should scenario-plan for city-by-city operating restrictions.
The gap between demo and deployment is structural, not temporary. Simulation-trained robots degrade from 95% lab accuracy to 60% in real environments. Ninety-five percent of construction AI pilots deliver zero measurable ROI. Sixty-four percent of digital twin projects stall before leaving pilot. These are not early-stage growing pains -- they reflect a fundamental difficulty in transferring controlled-environment performance to unstructured, variable real-world conditions.
Organizational readiness gates adoption more than technology. Across every mature practice -- warehouse robotics, digital twins, collaborative assembly -- the binding constraint is data governance, cross-system integration, and change management, not the robots or the AI. The MODEX 2026 survey confirmed it; so did Capgemini's finding that fewer than 25% of digital twin pilots reach operational deployment. Buying hardware without investing in the organizational infrastructure to support it produces expensive paperweights.
Labor shortages are real but do not automatically justify automation. The U.S. construction industry is short 499,000 workers annually. Japanese nursing homes cannot retain caregivers. German manufacturers face 6-month vacancy durations. Yet only a handful of robotic solutions have converted demographic pressure into repeatable, profitable deployment. The risk is that urgency drives premature investment in immature platforms -- particularly humanoids -- when narrow, proven solutions already exist for many of the tasks that matter most.
Go deeper: the full Physical AI & Robotics briefing -- the longer analytical write-up, plus every practice we track in this domain with its maturity rating, the tools to consider, and the evidence behind our assessment.