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 systems that navigate and operate in the physical world across road, air, marine, rail, and space. Heavily leading-edge: ADAS and warehouse robotics are good practice, but full autonomy in open environments remains elusive. Evenly split between advancing and stalled — regulatory frameworks and edge-case safety are the binding constraints. One of the largest domains with 22 practices spanning a wide maturity range.
The headline: Self-driving technology now pays its way wherever the operating area is bounded — mines, shipping lanes, rail networks, fixed delivery routes — and stalls everywhere it has to handle the open, unpredictable world. The frontier this fortnight moved on military and industrial deployments, not on cars that drive you home.
Autonomy has split into two markets. In bounded environments — a mine, a coastal shipping route, a metro line, a managed neighborhood — the machines work and increasingly make money; the obstacles left are cost, labor deals, and standards, not the technology. In the open world — your driveway, a busy unmapped city, all weather and all conditions — almost every confident timeline has slipped. Full self-driving (a car that handles every situation with no human) remains decades away. Conditional highway self-driving has technically arrived but is being abandoned by Mercedes, BMW, and Stellantis as too costly and legally murky. Robotaxis are the one open-world success genuinely scaling, led by Waymo in the US and Baidu in China, but both keep hitting walls — system-wide breakdowns, weather, and a public that refuses to trust them. If your business depends on bounded, repeatable routes, you are in the part of this domain that delivers; if it depends on open-road autonomy, you are betting on a timeline nobody can pin down.
Nothing was reclassified — and that steadiness is the news. After two years of constant up-and-down reratings, every practice held its position this cycle. The split between "works in a fenced area" and "still stuck in the open" hardened rather than shifted. Treat the bounded-environment use cases as investable today and the open-world ones as watch-and-wait.
The biggest new commitments came from defense and heavy industry. The US Navy committed over $4 billion for autonomous surface and undersea vessels; China deployed the world's largest all-electric autonomous mine-truck fleet; Japan put the first autonomous cargo ships into paid service. Capital is flowing to autonomy where the environment is controlled and the buyer is a government or industrial operator — not to consumer vehicles.
Tesla's credibility took two more hits. A Reuters investigation found its "10x safer" safety claim overstated roughly threefold, and the company was found to have quietly rewritten old purchase contracts to insert "supervised" language — effectively retiring its original full-autonomy promise. If a competitor or vendor cites Tesla autonomy numbers in a pitch, discount them.
A bus crash showed how literal the gap still is. An autonomous shuttle in Sweden was hit by a tram within an hour of its first paid trip because no one had programmed in the local rule that trams always have right of way. The lesson for any deployment: these systems only know the rules someone explicitly taught them.
New mandates land over the next 12 months. The EU's automatic-emergency-braking and driver-monitoring rules took effect 7 June; China's mandatory driver-monitoring standard arrives in 2027; the first global autonomous-shipping code and a US federal autonomous-trucking bill are advancing. If you run vehicles or fleets in these regions, confirm your suppliers are already on the new requirements rather than scrambling later.
Labor is organizing against automation, contractually. 45,000 dockworkers struck 36 US ports in May specifically to block port automation, and trucking unions are writing displacement protections directly into contracts. Any automation business case in logistics or transit now has to budget for negotiated labor terms, not just hardware and software.
Public trust is the ceiling no vendor has cracked. Despite robotaxis being measurably safer than human drivers, only 13 percent of US drivers are comfortable with self-driving, and acceptance collapses the moment the human steward leaves the vehicle. If customer-facing autonomy is on your roadmap, plan for a trust problem more data will not solve — and keep a visible human in the loop longer than the technology strictly requires.
The fenced area is the business model, not a phase you grow out of. Every operator making money — in delivery, trucking, robotaxis — did it by narrowing scope and staying there. Each attempt to break out into open cities or all weather has failed on cost, safety, or local bans. Assume bounded deployments deliver; treat "we'll expand to everywhere" as unproven.
Permission was the easy part; economics is the wall. Regulators have largely cleared the way, yet conditional self-driving still collapsed in the West on cost and liability, and a profitable autonomous bus in Japan was shut down the day its subsidy was cut. Regulatory approval tells you that you may deploy, not that anyone can make it pay.
Scale creates shared, catastrophic failure modes. One software fault stranded over 100 robotaxis in a Chinese city at once; a single spoofing attack froze an entire shuttle fleet. The more identical autonomous units you run on one system, the bigger a single flaw becomes — a risk that grows precisely as deployments succeed.
Go deeper: the full Autonomous Systems & Vehicles 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.