The science behind it
“A brain for the firm” is not a metaphor.
It is a research programme that runs back more than half a century. MAIA sits where three lines of it meet: the idea that a company is a system that can regulate itself, the idea that a company is a system that knows things, and the idea that a machine can reason in a way you can check — over a ledger that has been provable since 1494.
A note on honesty: below, we separate what MAIA directly builds on from the wider tradition it stands in. We name the second group as intellectual lineage, not as a claim that we implement their models.
Line one — directly built on
A company is a system that can regulate itself.
The study of control and communication — in machines, organisms and organisations alike — begins with Norbert Wiener (Cybernetics, 1948). W. Ross Ashby gave it a law MAIA takes literally: to stay in control of something, a regulator needs at least as much variety as the thing it regulates (An Introduction to Cybernetics, 1956; Design for a Brain, 1952). And Stafford Beer turned it into a model of the enterprise — the Viable System Model — in a book whose title we have borrowed on purpose: Brain of the Firm (1972; and The Heart of Enterprise, 1979).
MAIA operationalises that model: the layers that keep a firm viable — running the work, coordinating it, controlling it, auditing it, setting policy — become software, with a human sovereign above them (see below).
Line two — the tradition we stand in
A company is a system that knows things.
A second line treats the firm as an engine for processing information and decisions. Herbert Simon showed that organisations exist because real people decide under limits — the idea that became “bounded rationality” (Administrative Behavior, 1947; the term, Models of Man, 1957) — and argued that an organisation, like any designed artifact, can be engineered (The Sciences of the Artificial, 1969). Jay Galbraith framed organisation design itself as matching information-processing capacity to uncertainty (1973–74).
From there the question becomes knowledge: how a firm creates it (Nonaka & Takeuchi, The Knowledge-Creating Company, 1995), learns as a whole (Senge, The Fifth Discipline, 1990), and remembers past the people who leave (Walsh & Ungson on organizational memory, 1991). And it becomes cognition that is spread across people and their tools, not locked in one head (Hutchins, Cognition in the Wild, 1995; Wegner on transactive memory, 1986).
MAIA’s memory — the work, the files, the declared know-how and the record of what it proved — is a modern realisation of this idea. We cite it as lineage; the mechanism is ours.
Line three — directly built on
Reasoning you can check.
The third line is why any of this is trustworthy now. For decades, AI came in two flavours: formal, rule-based systems that could prove their conclusions but were brittle, and statistical systems that were flexible but could not explain themselves. The case for combining them — keeping the provable, explainable part and adding the adaptable part — has been argued for years (Gary Marcus, The Next Decade in AI, 2020; the neural-symbolic research community; Henry Kautz, The Third AI Summer, 2020; and DARPA’s “third wave” framing, Launchbury, 2017).
MAIA is built on exactly that combination: a formal, rule-based core that produces a checkable proof of what it did, paired with adaptable AI that proposes and explains. The proof idea has a precise pedigree in computer science — a computation can carry its own machine-checkable proof of correctness, verified by whoever relies on it (proof-carrying code; Necula, 1997). MAIA applies it to the place it matters most: every posting carries the proof of why it exists.
And the ledger itself is the oldest provable system in business. Double-entry bookkeeping — codified (not invented) by Luca Pacioli in 1494 — is a self-checking system: every transaction must balance, or the books declare themselves wrong. MAIA extends that lineage with tamper-evident, signed records (the cryptographic “triple-entry” idea; Grigg, 2005), auditing that runs continuously rather than once a year (Vasarhelyi & Halper, 1991; Groomer & Murthy, 1989), and checking real activity against the declared way things should run (process mining and conformance checking; van der Aalst).
What MAIA adds to the lineage
Two contributions of our own.
A human sovereign above the AI
The classic model tops out at policy. We add one layer higher: a person who sets the AI’s limits, can override any decision, and cannot be shut out. An AI that both runs a process and reports on its own health is a smoke detector wired to the arsonist; an independent human above it is the fix.
Proof-backed conformance
MAIA compares the way you declared a process should run, against what actually happened, against what was proven — and shows the difference as a checkable derivation, not a statistical guess about anomalies. Policy versus practice, with the receipts.
Why it beats the tools you already know
The same lineage, taken further.
- Process engines (Camunda-style). They fire a step when a box is ticked. MAIA proves each step, and the process is grown by real work rather than drawn in advance.
- Durable-execution platforms. Good at running code reliably for a long time; MAIA is people and AI over months, with memory and reasoning — not just durability.
- Case management. The closest relative in spirit, but with no cross-case learning and no proof trail. MAIA adds both.
- Process mining. It reconstructs a model from logs, statistically. MAIA checks against a model a human declared, so conformance is proven, not inferred.
- AI agent frameworks. There, the agents are the product. In MAIA the intelligence is one bounded, recorded step inside a durable process — not the thing you hand the business to.
The honest synthesis
Decades apart. Finally in one place.
Each of these ideas is old. They never came together into a product, because each was missing a piece: the vision of a self-regulating firm had no reasoning it could trust; the AI was either a black box or too brittle to use; and accounting’s provability stayed a manual, once-a-year affair. MAIA is where they meet — a firm that can reason about itself, in a way you can check, over a system of record that is auditable by construction. That is the whole claim, and it is a fact before it is a pitch.