The name is the architecture

IOIO — input and output

Tekkies know jackshit about information. They know about data. Data is what comes off a sensor, out of a database row, or down the wire from an API. Information is what data becomes when it has been typed, compared, and put into context with other data. Knowledge is what information becomes when it has been reasoned over against a goal.

The hierarchy is not new — Russell Ackoff named it in his 1989 paper “From Data to Wisdom,” and the field has called it the DIKW pyramid ever since: Data → Information → Knowledge → Wisdom. Each layer is a transformation, not a synonym.

The transformation from data to information to knowledge is the work that agents do. A Fundi reads typed objects out of the Object Store (data), compares them against other typed objects through the schema-as-data layer (information), and reasons over the comparison against a sovereign goal defined by the operator (knowledge).

The hand-off from knowledge to wisdom — the decision — stays with a human. Always. That distinction is load-bearing for everything else IOBOXX does, and it is the part of the AI hype cycle that the platform refuses to participate in.

BOXBOX — the black box

A black box in cybernetics is any system whose internal mechanism is opaque but whose inputs and outputs are observable. The term was formalised by W. Ross Ashby in An Introduction to Cybernetics (1956), which devoted an entire chapter to it. Ashby’s argument was that all real objects are in fact black boxes — complete knowledge of any system’s internal workings is impossible — and so the only honest way to study a system is through the protocol of input-output pairs observed over time.

The black box has a property that turns out to be the foundation of automation: given any two of {input, process, output}, the third can be derived. Give a process and an input, you can compute the output. Give an input and an output, you can infer the process. Give a process and an output, you can reconstruct the input that must have produced it. This is not a metaphor; it is the structural reason agentic AI works at all. Agentic AI runs the black-box derivation at the speed of light — across thousands of typed objects, against any of the three unknowns, continuously, and at marginal cost.

Most of the work inside any business is black-box derivation work. Match a tender item to a catalogue product. Reconcile an invoice to a purchase order. Classify a transaction. Rank a supplier. Surface an exception. Compute a margin. Detect a duplicate. Almost all of it can be automated. Almost all of it should be. Save the human decision.

IOBOXX is built for that reality. Every step of every BPMN-governed workflow has a principal — agent or human — and every transition is recorded in a commitlog that survives the workflow. The Draft → Review → Approved → Authorized lifecycle is not a UX flourish; it is the audit trail an accountable human relies on when the regulator knocks. See BPMN for how that principal model executes in production.

XXXX — the mesh of sovereign nodes

The XX is the part of the name the AI hype cycle most consistently ignores. It stands for the mesh of sovereign nodes — twins that own their own working memory, run their own agents, and share information with other twins only by explicit, permissioned grant.

From an invoice to a catalogue to a sensitive date of birth, every datum has a sovereign owner, every share is explicit, and every grant is auditable. It is a mesh, not a platform. No central control plane owns the data. No tenant becomes another tenant’s product.

Permissioned sharing is the part most “AI for business” pitches skip past, because their architecture cannot do it: their model lives in a single multi-tenant SaaS, their training data is everyone’s data, and their go-to-market is “give us your data and we’ll give you back insight.” That is the opposite of how regulated industries — and any business with a competitive moat — actually work. Real businesses share specific information, with specific counterparties, under specific contractual terms, for specific durations, with specific revocation rights. That is what XX delivers, by construction.

The mesh is what turns IOBOXX from “a memory system for a company” into a memory substrate for an economy. A supplier twin shares its catalogue with a buyer twin under one grant, its compliance certificates with a regulator twin under another, and its margin structure with no one. A clinical twin shares de-identified outcomes with a research twin while keeping patient identity inside its own sovereign boundary. None of this requires a central database, a federated query engine, or anyone’s “data lake.” It requires a grant, a typed reference, and a commitlog. See authorization for the grant model, and senses for the N → 1 / 1 → N topology that lets a sovereign node reach any system of record without point-to-point integration.

The name is the architecture

Read end to end:

• IO — input and output, the data → information → knowledge transformation that agents perform.
• BOX — the black-box derivation that lets agentic AI automate everything except the human decision.
• XX — the mesh of sovereign nodes with permissioned sharing of typed information across boundaries.

That is the platform. It is also the positioning statement. IOBOXX is the input/output black box of the agentic-AI economy, meshed across sovereign nodes, with the human decision preserved as the only thing that cannot be automated. Any time a visitor wants to know what IOBOXX is, the answer is in the name. The substrate behind it all is memory.