Every company that has tried to automate itself has run into the same wall, and it is not a technical one.
The tools are good now. An agent can read a contract, reconcile an invoice, draft the follow-up, update the record. Any one of those, demonstrated in isolation, is convincing. The demo always works. What does not work is the week after, when the thing has to run against a real process that nobody ever wrote down.
The gap is not capability
Businesses do not run on processes. They run on a hundred small agreements that live in people's heads: who signs off on what, what counts as urgent, which customer gets the exception, what we do when the usual route is closed. None of it is in the system of record. Most of it has never been said out loud.
Automation fails at exactly that boundary. Not because the model cannot do the task, but because the task, properly described, turns out to include forty things nobody thought to describe.
So the work is not building a better agent. It is making the undescribed part describable, which is slower, and which is mostly a human job.
What a platform has to be
We think a business automation platform is four things, and most of the market ships one of them.
It needs somewhere for the work to actually run, which is infrastructure. It needs the company's own context, the agreements and decisions and documents, in a form a machine can use. It needs agents that do the parts that should not be done by hand. And it needs people who can tell which is which, because that judgement does not come free with the software.
Ship only the agents and you get a demo. Ship only the infrastructure and you get a platform nobody knows what to do with. Ship only the people and you are a consultancy with a good story.
Why we build it in the open
The reason we run this as an ecosystem rather than a product is that the fourth part does not scale the way software does. You cannot hire enough people centrally to understand every company's undescribed half. What you can do is build the tools so that the people who already understand it can use them, wherever they happen to be.
That is why the products are separate and the community is not an afterthought. Product Engine runs the software. Tree holds the context. Agentic does the work that should not be manual. Mentors and the community are the people who know which forty things you forgot.
Where this actually is
Most of this is live and taking real money. Fulfilment across all of it is still substantially manual, and that is correct at this stage rather than a defect to design away. The constraint is not what we have built. It is that very little has yet been through a complete cycle: stranger, paid, delivered, came back.
We would rather say that plainly than describe a roadmap in the present tense.