To see why, look at services through three eras and a half.
Services 1.0 · human labor in person
Services 1.0 is the classic. Cleaning agencies, hotels, hospitals, taxis, restaurants, salons. Someone shows up and does the work. The modern face is Uber, but the shape is older than the wheel. Its defining feature is that it is bundled by default. You cannot rent a chef without renting the kitchen. You cannot book a hotel room without paying for the staff, the systems, the security, and the structure that holds it together.
The reason is structural. Humans do not unbundle cleanly. A worker is a single unit who arrives with their training, their judgment, and their physical presence fused. Services 1.0 scaled by adding workers, not by industrializing. Productivity gains came from logistics and software, never from the underlying labor.
Humans do not unbundle cleanly.
Services 2.0 · digital delivery
Services 2.0 took the parts of services that could go digital and rebuilt them in software. SaaS for business operations. Telemedicine for primary care. Online banking for retail finance. Streaming for entertainment. Where the work was already information, software took the human out of the loop.
This worked. The companies that built Services 2.0 are among the most valuable in the world. But Services 2.0 has a hard ceiling. It cannot pour the coffee, change the bedsheets, mop the floor, patrol the warehouse, install the equipment, or care for the elderly. The atoms layer of services stayed exactly where Services 1.0 left it.
Services 2.5 · vertical autonomy
Around 2018, a new category began to emerge. Call it Services 2.5. Vertically integrated, narrowly autonomous physical services. Waymo runs robotaxis in a handful of cities. Zipline delivers blood and medicine by drone. Skydio inspects infrastructure with autonomous flight. Nuro delivers groceries on dedicated vehicles. Each one solves a single physical service end-to-end, with the company owning the hardware, the brain, the routing, the maintenance, and the operations.
This was the right architecture for the moment. Foundation model brains were not ready. Hardware was expensive and bespoke. Remote operations infrastructure barely existed. The only way to ship a working autonomous service was to control every layer.
Services 2.5 proved two things. Physical work can be industrialized. And vertical integration has a ceiling. The moment any layer of the stack commoditizes, the bundle becomes less defensible. A service that built its own brain looks brittle when an open-source foundation model arrives that does the same job. A service that built its own hardware looks slow when Chinese OEMs ship comparable robots at a tenth of the cost.
Services 2.5 is a transitional architecture, not a destination.
The three forces of 2026
Three things are now happening at once that change the architecture.
The pieces are on the table. The question is not whether physical services automate. The question is what architecture wins.
The unbundling thesis
Look at how goods got here. In 1980, you bought from Sears or a department store that bundled selection, financing, delivery, and service into a single transaction. By 2000, specialty retailers had unbundled the categories. By 2010, Amazon had re-bundled the economy at the marketplace layer, with each product, each seller, each shipper competing on equal footing. Unbundling and re-bundling at the marketplace layer happened together. They compounded.
Services skipped the unbundling step because the underlying unit was a human, and humans came pre-bundled. With robots and remote operators, the unit unbundles. The skill is software. The hardware is a rental. The operator is global. Each layer prices, rates, and replaces independently. This is the architectural shift that Services 2.5 cannot make and that Services 3.0 demands.
Three SKUs
In Services 3.0, every service composes from three separately purchasable parts.
A skill subscription. Certified, royalty-bearing software that knows how to do a thing. Skills are written by domain experts, certified for use, and licensed per use or per period. The author earns royalties on every run.
Hardware rental, time-metered, by the second. The robot itself, with deployment, maintenance, and break-fix included. Hardware hosts own and operate the robots; consumers and businesses rent capacity. Robot depots, mobile depots, and on-site stations dispatch the right body to the right job.
Operator time. Certified remote humans who supervise, intervene, and provide the judgment that brains do not yet have. Operators are global, multi-timezone, rated per (skill, hardware) pair. As autonomy improves, operator hours per job decrease, but the role does not disappear.
These three SKUs price separately and recombine on demand. A buy box selects the winning combination on price, rating, and proximity. A real receipt looks like this:
Three suppliers, one transaction, transparent split. Every line is replaceable; every supplier competes; the buyer pays for what gets done.
The certification layer
The unbundle only works if the parts fit. A skill written for a Unitree humanoid does not run identically on a Figure humanoid. A skill certified on Physical Intelligence's π0.6 is not automatically certified on π0.7. Without a cross-vendor compatibility validator, the marketplace is a bazaar.
The certification layer is what turns a marketplace into an institution. Every (skill, hardware, brain) combination is validated before it goes live. Brain versions are pinned, so a model upgrade does not silently break a deployed service. Operators are certified per (skill, hardware) pair, with proctored evaluations. Every job in the marketplace tests one cell of a compatibility matrix, and that matrix compounds forever.
The certification layer is what turns a marketplace into an institution.
This layer is the equivalent of UL for appliances or DCI for digital cinema. It is the quiet infrastructure that turns a market into a category. It is also the most defensible asset on the platform, because every run adds a row no competitor can backfill.
Brain-agnostic by design
A common question: which brain does Services 3.0 bet on? The answer is every brain. Skild AI, Physical Intelligence, NVIDIA Isaac, open source models from Berkeley and Tsinghua, whatever ships next month. The marketplace does not pick winners at the brain layer. It lets brains compete for real-world jobs and routes the work to whichever combination delivers the buyer the best result. The brain companies are partners, not competitors. They make the engines. The marketplace is the shop.
Why now
The demographic forcing function is the closing argument.
Worker-to-elder ratios are collapsing in every G20 country. Services 1.0 cannot scale to meet that demand without immigration patterns no political system will accept. Services 2.0 does not solve physical work. Services 2.5 cannot scale horizontally fast enough.
Hardware costs are in free fall. Brain models are crossing the threshold. Operators are global. The window opens once, and the platform that owns the marketplace layer owns the next decade of services.
Robomyne
Robomyne is the first certified, unbundled marketplace for automation-native services. Three SKUs, five layers, brain-agnostic by design. The tagline is the spec: any brain, any skill, any robot, anywhere.
We are not building a brain. We are not building a robot. We are not building a vertically integrated service. We are building the marketplace where all of those compete on equal footing for the work that actually needs doing.
Services 1.0 was about presence. Services 2.0 was about software. Services 2.5 was about vertical autonomy. Services 3.0 has its own shape. Composable. Certified. Metered by the second. Anywhere it is needed.
The marketplace is what we are building.
Any brain. Any skill. Any robot. Anywhere.