LabsAI · Research Brief · Orchestration Intelligence
Introducing Structs
Language models reason through language. Orchestration must reason through systems.
August 17, 2026
A sentence is not a system. A task rarely exists alone — it has dependencies, permissions, participants, deadlines, data sources and compute requirements, and none of that structure is native to a language model’s medium. Structs — also written AI Structs — are the orchestration-native reasoning structures under research at Labsintelligence: the architectural units that would let an orchestrating intelligence understand systems, dependencies, routes, permissions, topologies, workflows, memory and state — and reason through them rather than about their descriptions.
Labsintelligence · lab1 of Labs In research & development at Labsintelligence
Three families, one loop. Reasoning decides what the system is. Learning improves the deciding. Control makes the decision safe to execute. The three Struct families divide along exactly those lines — and together they form the reasoning, learning and control loop an orchestration layer runs on.
01The three families
Systemic Reasoning StructsSRS · reasoning
The topological intelligence layer: decompose a request, identify boundaries, map dependencies, select orchestration paths, evaluate constraints, and update the picture as new information arrives. SRS is what lets a coordinating intelligence treat a system as a system rather than as a paragraph describing one.
Orchestrated Learning StructsOLS · learning
The layer that turns behavior into reusable structure: which sequences repeat, which coordinations succeed, how work actually moves. An orchestration layer should improve from the topology of work — not only from the text of its conversations.
Orchestrated Control StructsOCS · control
The execution and governance layer: converting a chosen orchestration path into controlled action — operating within permission boundaries, maintaining execution state, and keeping every step accountable. Without control, an orchestration is a plan; with it, the plan becomes an executed coordination.
02AI Structs and AI Orchs — the same stack, different jobs
The two names are easy to conflate and must not be. An AI Struct is a reasoning structure: it holds an understanding of the system — the dependencies, the permissions, the topology, the routes. An AI Orch is an execution structure: the live coordination object that binds participants, state and receipts into one operation with a lifecycle. A Struct never executes anything; an Orch never has to rediscover the system it runs in. If AI Structs help an orchestration layer reason about systems, AI Orchs are how it coordinates those systems into action.
And both sit around AI agents, not among them. The published hierarchy — models reason, agents act, Orchs coordinate — extends one level down: AI Structs understand the system an agent is about to act in; AI Orchs coordinate the agents acting; the agents themselves do the acting. An agent inside this stack inherits a map it did not have to draw and a coordination it did not have to negotiate — which is exactly what separates a governed agent from a loose one.
AI Structs reason about systems. AI Orchs coordinate them. Agents act inside them.
03What ships behaves as the loop — and only that is claimed
The Struct families themselves are in research and development at Labsintelligence. The wording rule this brief keeps is strict: shipped machinery is never called a Struct — but two shipped systems already behave as the loop’s poles, and they are the evidence the research stands on. The Studio’s offline improvement layer behaves as an OLS: it learns from recorded orchestration history under a promotion gate that vetoes any regression on held-out work — learning from how work moved, admitted only when strictly better. And TARG behaves as an OCS: six governance stages on every operation, execution state maintained, refusals on the record. The SRS pole is the openest question — reasoning natively over topology is precisely the capability the generative-transducer research programme exists to build.
Reasoning decides. Learning improves the deciding. Control makes the decision safe to execute.
The discipline all three families serve has its own note.
Contributing authors: Maya E. Davis · Duránd F. Davis Jr.
Tell us what you think, join us
This research is published while the questions are still open, and the systems it describes are live. We would love for you to join us — and please share your thoughts at research@labsintelligence.ai.
Citation
Please cite this work as:
Davis, Maya E., and Davis, Duránd F., Jr., “Introducing Structs.” LabsAI Research Briefs, Labsintelligence — lab1 of Labs Companies, Inc., August 2026.
Or use the BibTeX citation:
@article{labsintelligence2026introducingstructs,
author = {Davis, Maya E. and Davis, Duránd F., Jr.},
title = {Introducing Structs},
journal = {LabsAI Research Briefs},
publisher = {Labsintelligence, lab1 of Labs Companies, Inc.},
year = {2026},
month = {august},
url = {https://labsintelligence.ai/research/labsai/introducing-structs/},
}
