Outcome coordination is an architectural model in which AI agents express objectives and participating systems coordinate through explicit offers, commitments, evidence and settlement. Execution remains governed within each system, while cross-system obligations are represented as verifiable commitments rather than opaque chains of tool calls.
By Hammad Abbasi · Published on Zenodo · DOI: 10.5281/zenodo.21792756
AI agents are often placed inside the execution loop of enterprise and consumer software. The model selects tools, reconstructs parameters, carries intermediate state through context, responds to failures, and eventually causes an authenticated side effect. Each operation can be valid while the combined effect is outside the task the user intended.
This paper presents Outcome Coordination as an architectural response: organizing agent systems around explicit outcomes, bounded authority, decision points, evidence and responsibility for fulfillment. It separates work performed inside one organizational trust boundary from coordination that crosses between organizations, and shows how Lattice and Covenant Layer represent the two scopes.
A model in which agents express objectives while participating systems coordinate through explicit offers, commitments, evidence and settlement.
Individually permitted operations can still combine into an effect the user never intended once work spans multiple systems and organizations.
How a capability runtime keeps sequencing, state, credentials, failure policy and audit inside reviewed code, returning a projection the model can reason over.
How objectives, offers and authenticated acceptance turn cross-boundary work into verifiable commitments rather than opaque tool chains.
The public coordination objects and lifecycle that make authority, terms, acceptance, evidence and settlement explicit and inspectable.
Lattice governs execution inside one trust boundary; Covenant Layer coordinates commitments across boundaries. The two layers can be deployed independently or together.
Reference diagrams for direct-operation agents, the Outcome Coordination layers, Lattice execution flow, the Covenant lifecycle and the combined architecture.
Intent integrity, delegated authority, capability and provider execution, evidence and settlement, privacy, and dispute governance as assurance boundaries.
Open research on deriving the coordinate, revision and reauthorization, outcome equivalence, projection completeness, authority provenance and broker incentives.
Reframes the agent-facing interface from low-level tool calls to explicit objectives, offers, commitments, evidence and settlement, so agents interpret intent and apply policy rather than driving fulfillment step by step.
Distinguishes execution governed inside a single trust boundary from coordination that crosses organizational boundaries, keeping sequencing, credentials and failure policy inside reviewed code.
Defines objectives, offers, commitments, evidence and settlement as inspectable coordination objects with a clear lifecycle, making authority, terms and acceptance explicit and auditable.
Shifts responsibility for fulfillment to the systems that commit to an outcome, with evidence and settlement standing in for opaque, hard-to-audit tool chains.
Grounds the model in working code: Lattice as a capability runtime for structured execution within a system, and Covenant Layer as a protocol for commitment-based coordination between systems.
Abbasi, H. (2026). Outcome Coordination for AI Agents: Governing Execution Within Systems and Commitments Across Trust Boundaries. Working Paper, Version 1.2. Zenodo. https://doi.org/10.5281/zenodo.21792756
Version 1.2 — August 2026. First public release of the combined Lattice and Covenant Layer architecture. Each version is permanently archived on Zenodo under a concept DOI, with version-specific records linked to one another. Licensed under CC BY 4.0. View all versions on the Zenodo record.