- Working Group Name: Workflows and Process Integration
- Short Name / Acronym: [W&PI]
- Date Approved: 2026-05-13
- Last Updated: 2026-07-23
- Homepage / Repo: https://github.com/aaif/wg-workflows-and-process-integration
- Primary Contact (Chair): Yaron Schneider, yaron@diagrid.io
- Secondary Contact (Co-Chair): Adam Seligman, adam@workato.com
The Workflows and Process Integration Working Group advances the Agentic AI Foundation’s mission by establishing a shared workflow model and common terminology for agentic AI systems, extending open specifications and projects, producing interoperability guidelines and design patterns that allow agent workflows to operate across frameworks, tools, and execution environments.
This Working Group was formed to address:
- How do we define common workflow primitives (tasks, tool calls, branching, handoffs) that work across agent frameworks and runtimes?
- How do we standardize the way agents coordinate with external systems, tools, and human reviewers within multi-step workflows?
- How do we ensure agent workflow definitions are portable across different frameworks and execution environments, avoiding vendor lock-in?
- How do we enable adoption of agent workflows that require stateful replayability and reliable failure recovery?
- How do we establish production-ready operational patterns for running agent workflows at scale?
The work of this WG supports:
- Support the primary imperative that agentic AI requires complex, multi-step interaction by defining how workflows and agents orchestrate each other, including existing systems and human reviewers across organizational boundaries.
- Lower enterprise adoption barriers by establishing common patterns and terminology for workflow execution, failure recovery, human-in-the-loop coordination, and production operations.
- Advance agentic AI technologies by producing vendor-neutral workflow specifications and reference architectures that enable interoperability across frameworks and execution environments.
A. Agent Workflow Models and Semantics
- Common terminology and primitives for agent workflows (tasks, tool calls, branching, sub-agents, retries etc.)
- Standard representation of agent workflow structures across frameworks
- Execution semantics for agent workflows (step ordering, parallelism, conditional paths)
- Agent handoff semantics between agents and sub-agents
- Motivation: Industry adoption challenges across agent frameworks due to misalignment on key definitions and lack of knowledge in translating deterministic workflow models to non-deterministic models used by autonomous agents
B. Long-Running and Stateful Agent Execution
- Patterns for workflows that run for minutes, hours, or days
- Handling partial completion and resuming interrupted agent workflows
- Workflow state persistence models
- Handling retries, idempotency, and failure recovery in agent workflows
- Workflow-level audit logs and systems of record (execution history, step outcomes, decision points) — distinct from runtime observability signals such as traces and metrics, which are handled by the Observability Working Group
- Motivation: Enterprise adoption challenges with long-running AI agents that are sensitive to data loss. Lack of controls and systems to deal with interruptions in agent workflows that can cause loss of state, context and progress
C. Tool Invocation and External System Coordination
- Standard workflow patterns for agent interaction with external systems
- Workflow coordination between agents and tools (APIs, CLIs, services)
- Handling multi-step tool execution within workflows
- Consistency patterns when agent workflows interact with external stateful systems
- Motivation: Enterprise workflow orchestration challenges related to connecting multiple workflows into a single, coherent execution model that can be governed centrally and run reliably
D. Human-in-the-Loop Workflow Patterns
- Workflow patterns for human approvals and interventions
- Interruptible workflows that pause for human review
- Escalation and fallback workflows when agent execution cannot proceed autonomously
- Design patterns for combining autonomous and human-driven steps
- Motivation: Enterprise agent adoption requirements for gatekeeping agentic operations with human review and approvals
E. Workflow Portability and Interoperability
- Portable workflow descriptions across agent runtimes
- Workflow interchange formats
- Agent handoff between different frameworks or execution environments
- Compatibility guidelines for workflow portability
- Motivation: Enterprise environments will use multiple frameworks and need a standard way to communicate and share state across multiple technology stacks
F. Operational Patterns for Production Agent Workflows
- Reference architectures for running agent workflows in production systems
- Failure handling patterns and workflow lifecycle management
- Best practices for operating and scaling differnet agentic workflows (deterministic, probablistic, etc.)
- Motivation: Enable platform and infrastructure teams to run workflows at their desired scale and understand how to create reusable patterns for workflow deployments across the organization
- Model training, LLM architecture, and model development are outside the scope of this working group.
- Prompt engineering methodologies and prompt optimization techniques are deferred to other technical groups.
- Security, privacy, and attack surface analysis are handled by the Security Working Group.
- Observability, tracing, debugging, and telemetry for agent systems are handled by the Observability Working Group.
- Ethics, governance frameworks, and regulatory considerations belong to the Governance & Risk Working Group.
- Evaluation of model reasoning quality and accuracy belongs to the Accuracy and Reliability Working Group.
- Assumptions: Agent frameworks and runtimes will continue to evolve independently; this WG produces standards and patterns that layer on top of existing execution environments rather than replacing them.
- Dependencies: This WG has interactions with the Reliability & Accuracy Working Group (reliability expectations of agentic workflows), the Observability Working Group (tracing and telemetry for workflow runs), and the Identity & Trust Working Group (agent identity and delegation within workflows).
- Define a taxonomy of common workflow terms and primitives used across agent frameworks
- Identify and document a prioritized list of use cases for agent workflows across enterprise and consumer contexts
- Produce a workflow reference architecture for agent workflow execution
- Workflow Taxonomy — Owner: WG, Format: document, Target: 08.29.2026
- Workflow Best Practices — Owner: WG, Format: document, Target: 10.29.2026
- Workflow Reference Architecture — Owner: WG, report: document, Target: 08.29.2026
- Horizontal Use Cases Landscape — Owner: WG, Format: report, Target: 09.29.2026
A deliverable is considered complete when:
- Presented to the broader WG membership, with feedback incorporated or explicitly dispositioned
- Reviewed and approved via a WG super majority vote with no unresolved objections
- Published in the WG repository in its final form
- Timeliness: Deliverable targets and milestones met
- Community: Active contributors, meeting attendance, participation from 5+ member organizations
- Adoption: 5+ references and/or attributions from end-user companies, downstream projects or frameworks
- Consensus-driven with chair-led facilitation; voting as fallback
- Work tracked in: GitHub repository
- Primary artifacts: best practice guides, reference architectures, whitepapers, design pattern catalogs
- Secondary stretch-goal artifacts: Specifications
- Cadence: Every other week; Thursday 8:00am PT
- Duration: 60 minutes
- Time Zone Considerations: None
- Open Meetings: Participation is open to all individuals and organizations consistent with AAIF and Linux Foundation policies.
- Minutes/Notes: Google Drive (shared with WG members upon joining)
- Recordings: Available via LFX OpenProfile. Create a Linux Foundation (LFX) account to access meetings, AI summaries, recordings, and more.
- Zoom: Meeting link shared with WG members upon joining
- Async: Mailing list, Private Discord channel (link shared with WG members upon joining)
- Sync: Zoom (see meeting link above)
- Announcements: Mailing list
Participation is open to all individuals and organizations consistent with AAIF and Linux Foundation policies.
- Contributors: any WG member attending meetings or contributing asynchronously.
- Chairs/Co-Chairs: individuals responsible for operations and facilitation.
To join, a participant should: join the mailing list, sign the CLA/DCO if required, and request an invitation to the Working Group.
- Follow the Code of Conduct and collaboration norms.
- Make contributions in the open (issues/PRs) whenever possible.
- Declare conflicts of interest when relevant.
- Chair(s): Yaron Schneider
- Co-Chair(s): Adam Seligman
- Program Manager: [TBD]
- Chairs are selected by: Election
- Term Length: 12 months
- Renewal: Allowed, maximum of 2 consecutive terms. Positions carry on to next term automatically if no one else is contending.
- Removal: Super majority vote
- Default method: rough consensus documented in issues/meeting notes.
- When consensus cannot be reached:
- Vote rules (if used): quorum 50%, threshold simple or super majority, voting eligibility: Platinum and Gold members.
- Escalation path: TC / Foundation Governing Board
Quorum is met when 50% eligible voters are present or 50% have responded asynchronously.
- Liaison(s) to other WGs: [TBD]
- Shared deliverables/dependencies:
- Reliability & Accuracy Working Group: reliability expectations for agent workflows
- Observability Working Group: tracing, telemetry, and debugging for agent workflows
- Identity & Trust Working Group: agent identity, delegation, and authorization within workflows
- Standards bodies / industry groups: [TBD]
- Upstream/downstream projects: [TBD]
- Policy for external statements: [TBD]
- Code contributions: [e.g., Apache-2.0/MIT] (or "per repository license")
- Documentation/specs: [e.g., CC-BY-4.0] (or "per repository license")
- Meetings and communications must follow the foundation's antitrust guidelines.
- Avoid discussions of pricing, market allocation, or other restricted topics.
This WG adheres to the Linux Foundation Project's Code of Conduct.
- Expected cadence: As needed, with quarterly progress reviews
- Versioning scheme: SemVer for specifications, date-based for reports and guidance documents
Chair or co-chair sign-off required.
- Deprecation policy: Deliverables may be deprecated by a WG super majority vote with a minimum 30-day notice period
- Sunset criteria: No active maintenance or community interest for 6 months
This charter may be amended by:
Super majority vote of the Working Group and documentation in the WG repository. Subject to TC/board approval if required.
By approving this charter, the Working Group commits to operating transparently, in the open, and in alignment with foundation policies.
- Approved By: [TC]
- Date: 2026-05-13
Runs meetings, sets agendas, ensures notes, drives milestones, represents WG in cross-WG coordination.
Provides substantive work items (PRs/docs/issues), participates in reviews and discussions.