The self-driving company, in production.
Business automation used to mean quarters of consultants modeling workflows. Tai42 introduces the self-driving company — business functions that run themselves as production applications, humans at the named decisions. The industry's own evidence says raw autonomy fails and structured autonomy works — in Carnegie Mellon's benchmark of real workplace tasks the best agents completed under a third; MIT found most enterprise pilots never touched the P&L; Anthropic's Project Vend turned a loss into a profit purely by adding oversight structure. That is why, in the applications we build, the AI has no tools and humans sit at the named decisions: we sell the structure, and we hold ourselves to the same standard.
How the company is built — three layers
The open-source runtime.
The engine underneath everything we do is public: read the code, run it on your own infrastructure, extend it. It exists so that a small company can be deep tech an enterprise can rely on — you don't have to take our word for what the engine does. Permissions, audit, and delegated access are part of the open core, because governance is the product, not the paywall. Tai42 builds and runs its business on this runtime; the code is open; the company sells the hosted platform and enterprise layer on top — never a different core.
BabelDag, the platform.
Built on that runtime and operated by us: the deterministic flow engine, the visual builder, the governance layer, and the optimizer that watches live behavior and compiles it into fast, cheap, deterministic flows. Delivered as a private, dedicated tenant your team operates — your keys, any model, your data in your tenant. One rule makes it safe: the AI has no tools. Models write, interpret, and propose; only flows touch data, systems, and money.
The engagements.
We take your working demo to a production application — the readiness review, the gate, the build, acceptance you verify, a license only if you keep it. Deterministic where money moves, AI where judgment is needed, humans at a named handful of decisions. What you keep is a running application on your tenant, your flows as configuration, and us on call for the hard parts.
Why the layers are separate. So each one can make you a promise the others don't depend on. You can verify the engine without trusting us — the code is public. You can run your operations without sharing them — your tenant is private. And you can start without committing — engagements are short and end at gates you verify. Three layers, three independent guarantees.
How we hold ourselves to it
Every function we automate for a client, we run inside tai42 — our product management, our operations, our own company. Our org chart lists our agents next to our humans, each agent with a human owner. We publish no number we haven't measured on a live delivery. We claim no certification we can't hand over. And every role here, including the founders', ends with the same clause: …until a flow can do it. When we say production-grade, we mean: we live here.