Here is the heresy I want to argue today, and I want to argue it as someone who has spent a year writing about humanoid robots and another twenty before that watching software industries reshape themselves: the software running your robot should not belong to the company that sold you the robot. Not exclusively. Not by default. And not for much longer.
Hardware will always be the hard part. Actuators, harmonic reducers, force-torque sensors, battery packs that survive a 10-hour patrol — none of that is going to be democratised. The factory line that puts a humanoid together involves capital expenditure that small teams cannot replicate. I am not arguing with that. I am arguing about everything else — the layer that turns a beautifully engineered chassis into your robot, doing what you want it to do, in your home or workshop or care facility. That layer no longer needs a thousand-engineer R&D department to build. And the industry has not caught up to that fact.
The Cost of Software Just Collapsed
The Pickurai team published a piece recently called "Vibe Code: The New Social Elevator" that I have been thinking about ever since. Their argument, in a sentence, is that writing code has just become absurdly cheap — not because programmers got faster, but because the act of programming was rebuilt around natural-language collaboration with an AI. People who have never written a function before are now shipping working software. Senior developers are producing in a weekend what used to take a sprint. The economic floor of "what is worth building" has dropped through the basement.
Now apply that to robotics. The robot industry is still operating on a software model from 2015: every manufacturer ships a closed firmware stack, a locked SDK with gatekeeping and approval cycles, an app store that they curate, and a roadmap that moves at the speed of a corporate quarterly review. They do this because, historically, writing capable robot software was extraordinarily expensive — so the company that paid for it was entitled to control it.
That premise has just evaporated. The cost of a good developer plus a capable AI coding assistant in 2026 is not a fraction of what a robot manufacturer's internal team costs — it is a fraction of a fraction. And the pool of people who can now contribute meaningful robot software has expanded by an order of magnitude. Continuing to gatekeep the software layer in this environment is not strategy. It is denial.
Hardware Open. Software Open. Two Different Conversations.
Let me be precise, because the open-source debate in robotics tends to collapse into a confused argument about whether anyone could build their own Atlas in a garage. Nobody is arguing that. The mechatronic stack — the joints, the spine, the sensor array, the thermal management, the safety-rated motor controllers — is exactly the kind of thing where centralised manufacturing, big-budget R&D and tight integration genuinely earn their keep. Manufacturers should be paid handsomely to do that work well. Hardware is non-negotiable.
The software layer is a different animal. It is the part where personalisation, experimentation, and a thousand niche use cases live. It is the part where a single developer in Lisbon, working on a Saturday, can produce a behaviour for a humanoid that the manufacturer's internal team would never bother to ship — because it serves a hundred users, not a hundred thousand. The economics of who-should-write-this are flipped. And the only thing standing in the way is the artificial wall that says only the manufacturer is allowed to touch this.
"Hardware is the body. Software is the personality. There is no good reason that every robot of a given model should be forced to share the same personality."
What Users Actually Lose When Software Stays Closed
The most obvious cost of closed robot software is the one users experience every day and rarely articulate: your robot is configured for the lowest common denominator of its market. The dance routines, the conversation style, the patrol behaviours, the way it greets your kids in the morning — all of it was designed to be acceptable to the median buyer in the manufacturer's largest market. Which means it is, by construction, not actually right for you.
An elderly person who lives alone and wants their companion robot to remind them to take medication on a non-standard schedule. A small clinic that wants their patrol robot to recognise the specific layout of their corridors. A workshop that wants its arm to perform a sequence the manufacturer never anticipated. A parent who wants a story-time routine in their own dialect. Each of these is a perfectly reasonable request that the manufacturer will never fulfil — because the engineering cost of shipping it for one user does not pencil out at corporate scale.
But it pencils out beautifully for an indie developer, or a small studio, or even the user themselves. That is the whole point of the vibe-coding revolution: tasks that used to require an engineering team can now be completed by one motivated person in an afternoon. The ceiling on personalisation just got lifted. The wall that prevents anyone from reaching it is the closed software stack.
What Manufacturers Stand to Gain (Yes, Gain)
The instinct of every robot company I have ever covered is to treat their software as a moat. I want to make the case that, in the post-vibe-coding economy, the moat is now the swamp — and the smart move is to drain it deliberately.
A community-maintained software stack offloads enormous engineering hours from the manufacturer's payroll. The savings can be redirected to where they actually matter — the hardware that nobody else can build.
A thousand independent developers shipping behaviours in parallel will out-iterate any internal team. The robot becomes more capable every week without the manufacturer touching a line of code.
The platform plus a 70/30 revenue split with developers turns a hardware sale into a recurring marketplace. Apple did not get rich because it wrote every iPhone app — it got rich because it stopped trying.
When something goes wrong with closed firmware, the manufacturer owns the failure. When the community ships a buggy module, the platform survives — and the open-review process is, in practice, more rigorous than any internal QA cycle.
University labs and indie researchers will publish work on platforms they can actually modify. Closed stacks see almost no academic activity. Open stacks get a permanent firehose of free innovation.
Less in-house software headcount means lower fixed costs per unit. That margin can be passed on. The cheapest robots on the market will be the ones whose manufacturers stopped pretending to be software companies.
The Indie and SME Opportunity
Here is the part that genuinely excites me. If a manufacturer like Unitree, Booster or Engine AI announced tomorrow that the software stack on their next model was open and the SDK was permissive, what happens next is not chaos. What happens next is exactly what happened with the iPhone in 2008, with WordPress in 2005, with Android in 2010 — a Cambrian explosion of small businesses built around the platform.
A two-person studio in Madrid builds a behaviour pack for elderly care. A solo developer in São Paulo writes the definitive interaction module for autistic children. A small Korean company specialises in workshop-floor routines for family-owned manufacturing. None of these would ever be greenlit inside a robot manufacturer. All of them are perfectly viable as standalone businesses on top of an open platform — and each one makes the underlying robot more valuable to buy in the first place.
This is the same dynamic that made the smartphone the dominant computing object of the last fifteen years. It was not the hardware specs. It was the third-party software ecosystem that the hardware made possible. Robotics is staring at the same inflection point and most manufacturers are still acting like the keepers of a walled garden, several years after the wall stopped serving any purpose.
The Counter-Arguments, And Why They Mostly Don't Hold
"What about safety?" A real concern, and the answer is the same one Linux figured out three decades ago: a tiered architecture, with safety-critical motion control kept in a signed, manufacturer-maintained kernel, and the personality and behaviour layers above it open to the community. ROS already does something close to this. The pattern works.
"What about liability?" A robot running community-written software is a robot whose owner has chosen to install community-written software. The same legal framework that covers a custom Android ROM, a third-party drone flight controller, or a self-modified power tool already covers this. Liability is solvable. It is just not yet written down.
"What about competitive advantage?" The manufacturer that opens their software stack first, and does it with a developer-friendly licence and a genuine revenue-share marketplace, will define the standard for the entire industry — exactly as Android did. The competitive advantage of being the platform is far, far larger than the competitive advantage of being one walled garden among many.
What I Want to See in the Next Twelve Months
I am not asking every humanoid maker to abandon their entire firmware tomorrow. I am asking for three concrete moves, and I think the first manufacturer to make them wins the decade.
Open the SDK without gatekeeping. No approval cycle, no NDA, no "developer programme" with quarterly fees. Publish the API, the simulator, the safety boundaries, and let people build. The cost of doing this is zero. The number of robot software developers in the world doubles overnight.
Open the high-level behaviour layer. Movement, conversation, task scheduling, multi-modal perception above the safety kernel — all of it open source, on a permissive licence, with a clear contribution process. Keep the motor-control layer signed and manufacturer-maintained. Let the community own the layers where personalisation actually happens.
Run the marketplace, take the cut. Ship a first-party app store where indie developers and SMEs publish behaviours and modules. Take a fair revenue share. Curate for safety, not for taste. The platform earns money from every transaction without writing the software being sold. This is the single most-tested business model in modern technology, and somehow no robot manufacturer has copied it yet.
The Verdict: This Is Not a Question of If, Only of Who First
Closed robot software made sense in a world where writing software was expensive and writing robot software was the most expensive software of all. That world ended somewhere between the release of the first capable AI coding assistants and the moment Pickurai's writers correctly identified vibe coding as a social elevator rather than a curiosity. The cost curve has crossed. The old gatekeeping is now leaving money on the table — for the manufacturers, for the developers, and most of all for the people who actually own these robots and are stuck with whatever personality their factory shipped from across the world.
The hardware will keep belonging to the companies that can afford to build it. That is fair. That is right. But the software, the part that decides what your robot actually is when it walks through your front door — that should belong to whoever can write it best, on whatever schedule serves the people who will use it. In the year 2026, that is no longer a corporation. It is a network. And the first robot brand to admit that openly will be the one we are all still talking about in 2036.
Hardware, non-negotiable. Software, for the people.