LabsAI · Research Brief · Orchestration Intelligence
Introducing the Orchestration Graph
A conversation history remembers words. An orchestration graph remembers work.
August 17, 2026
Every AI product keeps a transcript. Almost none keeps a record of what was actually done — which actions were attempted, which of them verifiably landed, which orchestrations ran, what was refused and why. The Orchestration Graph (TOG) is that record: the persistent map of how intelligence moved through a session, kept as data rather than prose, and belonging as much to the person as to the system. It is the trust layer of Orchestration Intelligence — the difference between a system you have to believe and a system you can check.
Labsintelligence · lab1 of Labs Words are what was said · the graph is what was done
Why a graph and not a log. A log is what one component wrote down. The graph is the session’s work as one connected record: a spoken instruction — a Speech‑to‑Action or Speech‑to‑Orchestration Voice Command — becomes a tool operation, the operation becomes an outcome, the outcome becomes a receipt, and the receipts of many participants connect into the shape of the coordination itself.
01Receipts are the atoms
Four kinds of record compose the graph:
Tool events
every operation the intelligence ran — named, timed, and marked for whether it succeeded.
Action receipts
what happened on the surface, with the outcome checked against the surface itself rather than assumed.
Orchestration lifecycles
each AI Orch’s transitions — created, delegated, paused, resumed, forked, reconciled, completed — receipt by receipt.
Governance decisions
every refusal the TARG gate recorded, with the stage that objected.
Action receipts carry a five-valued outcome — and the vocabulary carries the discipline:
verifiedattemptedblockedfailedheld
One spoken instruction as the graph records it: three operations, their outcomes, an orchestration reconciled across participants, and the receipts that survive as the exportable trace.
Verified means seen on the surface, not assumed. Attempted means posted but not confirmed — and the intelligence is told to claim no more than that. The graph inherits the same truthfulness discipline the Interactive Intelligence Benchmark weights heaviest.
02The record is the person’s too
A trust layer that only the operator can read is telemetry. The graph is surfaced where the person is: the Studio’s Work view renders the session’s record as it builds — actions with their outcomes, orchestrations with their lifecycles, refusals with their reasons — beside the Sessions view that keeps the words. And the record travels: one tap exports it as an A2F .orch document, the append-only trace format, archive profile — a file of evidence, not an executable.
A conversation history remembers words. An orchestration graph remembers work.
03The shape of a session
Rendered along a timeline, a session’s graph reads like a strip of decisions — and the outcome vocabulary is visible at a glance. Green marks are verified work; hollow marks are attempts honestly labelled; red marks are the refusals and failures a transcript would never show you:
The same discipline the benchmark scores from the outside, visible from the inside: a held mark is a confirmation the person granted; a blocked mark is governance on the record.
04What the graph makes possible
The first generation runs in LabsAI Studio: per-session, receipt-complete, visible, exportable. What the graph reaches toward is in research and development at Labsintelligence: persistence across sessions into a genuine topology of work — which routes are effective, which coordinations repeat, which patterns deserve to become reusable structure — so the system improves from the shape of work rather than only the text of conversations. A transcript can teach a system what people say. Only a graph can teach it how work actually moves.
The trace the graph exports has a specification of its own.
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 the Orchestration Graph.” LabsAI Research Briefs, Labsintelligence — lab1 of Labs Companies, Inc., August 2026.
Or use the BibTeX citation:
@article{labsintelligence2026introducingtheorchestrationg,
author = {Davis, Maya E. and Davis, Duránd F., Jr.},
title = {Introducing the Orchestration Graph},
journal = {LabsAI Research Briefs},
publisher = {Labsintelligence, lab1 of Labs Companies, Inc.},
year = {2026},
month = {august},
url = {https://labsintelligence.ai/research/labsai/introducing-the-orchestration-graph/},
}
