Poesis
Define the world in software; generate it.
A thesis that refused to stay prose.
Every organization’s definitions, its DNA — architectures, processes, decisions, standards, contracts, responsibilities — live in prose, where they decay and automation stops. Poesis, with generative AI, is designed to end that. A definition — the generative precondition — stops being an unmaintainable document and becomes a typed, versioned object carrying its own semantics: structured, executable, verifiable, continuously reconciled with the system it defines — enabling the generative AI that acts on it to maintain systems autonomously. The thesis was published as open research before it became an open-source stack.
Origin
It began with a blocker, not a product.
Poesis’s founders spent their careers inside large IT estates: architecting them, governing them, selling and scaling the platforms they run on. Every estate ran on definitions — architectures, obligations, processes, constraints. And every estate kept them in prose. Wikis, slide decks, compliance spreadsheets, all decaying from the day they were written.
The deeper observation was about automation. Prose is not just where definitions decay; it is where automation stops. A definition in a document can be read, debated, and forgotten. It cannot be executed, evaluated, or regenerated. Every automated system downstream needs a human to re-interpret the prose into code, configuration, and tickets, so the loop a viable system depends on — act, observe, correct — breaks at that boundary. Prose blocks autopoiesis: a system’s capacity to produce and renew itself.
Then generative AI made production cheap. Code, configuration, whole systems, brought forth on demand. The blocker turned existential: systems now produce the components that produce them, while the organization they must conserve still lives in documents nothing can act on. Out of that tension came the research. Out of the research, the project.
The research
The argument the enterprise is built on.
The position was built as a chain of published research notes, each establishing a step the stack now embodies. Read in order, they are the company’s reasoning.
The inversion — where the intent comes from
Classical systemics describes systems after the fact, so governance audits the past. The research began by inverting that.
- The generative inversion — from describing systems to defining them
Make the definition primary and generative — the definition generates and governs the running thing, and the system becomes self-correcting rather than merely documented.
- Definition versus description: an opposition that may not exist
The old opposition dissolves: definition makes observation intelligible; description makes definition corrigible. The regulation loop systemics always drew becomes operable.
- Standardizing the THINK layer of IT
OpenTelemetry gave the RUN layer a neutral standard; the definitions those runtimes enforce — the THINK layer — still have none. That missing standard is what GSM specifies.
The generative era — why now
Generative AI did not create the definitional problem. It made ignoring it untenable — and solving it tractable.
- Generative AI is as probabilistic as human intelligence
Probabilistic reasoning is not a defect to repair — human intelligence is probabilistic too. Humanity’s remedy was never determinism; it was methodology and institutions. Machine reasoners should inherit both.
- Operational AI governance: the Poesis position — governance owns the concerns
A model can reason about a concern but cannot own one. Concerns are owned by stakeholders and carried by governance: the model proposes; the stack and the human dispose.
- Better context beats bigger models: why Poesis matters for language models
Do not ask the model to carry the organization in its weights. Governed context — which definition is in effect, who holds authority, what is permitted — beats bigger models for bounded enterprise work.
The name — what it commits us to
Poiesis (ποίησις) is the Greek word for bringing-forth. Maturana and Varela joined it to auto — self — and named the property that distinguishes living systems: producing the components that produce it.
- Governed synthetic autopoiesis — why Poesis is called Poesis
Generative AI made self-production cheap. What it did not make is the conserved organization that keeps self-production from becoming dissolution — the gap Poesis is named for.
- The Poesis stack as a research system
The constitution generalizes beyond IT: open-ended research is not a workflow to accelerate but an institution to constitute. The first domain is a starting point, not a boundary.
The thesis
Definitions belong in software, not in prose.
The notes converge on one thesis. An organization’s definitions must be typed, versioned, machine-operable objects: primary and generative, realized into systems that produce evidence, evidence that corrects the definitions. Not a model drawn after the fact. A loop operated continuously.
The generative era sets the terms of that loop. Probabilistic reasoners, human or synthetic, are not made deterministic; they are governed. Concerns stay owned by human stakeholders and carried by explicit governance. The model proposes against governed context; deterministic evaluation and human disposition decide what becomes action.
Held together, these commitments name what the enterprise exists to make real: governed synthetic autopoiesis — systems that continuously produce and renew their own organization without escaping the definitions, constraints, and human purposes that govern them. The theory is not ours to invent; it is systemics, from Beer’s Viable System Model and von Bertalanffy’s General System Theory to Wiener and Ashby’s cybernetics and Maturana and Varela’s autopoiesis. Our contribution is making that lineage executable.
The DNA
What the thesis defines.
The thesis took organizational form. The stack, the domains, and the way Poesis runs itself are not adjacent to the research — they are what it defines.
Definitions that machines must act on need a shared, vendor-neutral form. GSM is that THINK-layer standard: typed, versioned, evaluable, published openly and prepared for neutral stewardship.
Definition generates, realization produces evidence, evidence corrects the definition. SIE is that loop running: continuous evaluation and reconciliation of the real world against its governed definitions.
Humanity never made its reasoners deterministic; it built institutions around them. ITIP and SAF are those institutions for machine reasoners — the model proposes, humans dispose, in the first domain and in delivery itself.
Poesis is first an open-source initiative: the GSM standard is published openly, the SAF framework is Apache-2.0, and the engine is source-available on a scheduled path to open. A stack that claims to carry an organization’s definitions must be inspectable by the organizations asked to trust it.
An enterprise claiming governed self-production must run on it. Every Poesis repository — engine, applications, these sites — is defined and governed on Poesis’s own stack, inspectable on GitHub. Autopoiesis, applied first to ourselves.
Leadership
Founded by practitioners, not observers.
Poesis was founded on a thesis its founders had already lived: careers spent governing, building, and selling enterprise systems — watching the definitions those systems depend on decay in documents, and every automation stall at the prose boundary. The company exists to move the definitions inside the loop.
Clément Cazaud
Chief Executive, Product & Architecture Officer
Enterprise & Solution Architect, Software Engineer, AI Agentic Expert & AI Transformer, Serial Systemizer, and Entrepreneurial Engineer. Clément has spent his career turning fragmented, brittle IT landscapes into coherent, evolvable systems — driven by a lifelong instinct to systematize whatever he works on — or in. GSM and the Poesis stack are that instinct, generalized — made tractable by the generative AI era.
Brahim Ben Helal
Chief Growth Officer
A career at the intersection of enterprise software and commercial strategy — SaaS, AI, digital trust, and enterprise sales. Brahim designs the go-to-market systems that turn complex platforms into adopted ones. At Poesis he leads growth, ecosystem development, and customer engagement — keeping the platform accountable to real enterprise problems, not just to its own architecture.
Hamza Abidi
Chief Technology & Engineering Officer
Hamza is one of those rare minds that make hard things look simple. Razor-sharp logic, instant comprehension, and a natural gift for absorbing complexity let him adapt to any terrain and master it as if he had always been there. That is why he owns the technical foundations at Poesis — because when the ambition is as large as Poesis’s, the foundations must be trusted to someone who has never met a ceiling.