Methodology: How This Reference Is Built and Maintained
AgenticSwimlanes.com is an independent, vendor-neutral reference for AI agent workflow diagrams. Two citation disciplines run in parallel: every swimlane on the site references a real, public, named source, and every BPMN-shaped diagram is checked against the OMG BPMN 2.0 specification.
What this site is for
The reader is a process owner, BPMN practitioner, ops architect, or automation lead who needs a credible reference that survives review by a BPMN-aware colleague. The site is built so the answer to “where did you get this diagram?” is always a named source.
The site is not a tutorial for building agents (that is the territory of buildingeffectiveagents.com). It is not a definitional reference for AI agents (that is whatisanaiagent.com). It is not a workforce-impact calculator (that is aijobimpactcalculator.com). It is not the structural / org-chart view (that is agenticorgchart.com).
The citation discipline
Every swimlane on this site references a real, public, named source. The hand-built SVGs cite the source paper, framework documentation, or named case study at first appearance. The BPMN-correct per-process diagrams carry the source citation in the figcaption with the access date. Where the source cannot be verified, the example is excluded from the site.
Source pool used across the site:
- OMG BPMN 2.0 specification. omg.org / spec / BPMN / 2.0 (formal/13-12-09, accessed April 2026). Normative reference for shape semantics. Adopted as ISO/IEC 19510:2013.
- Anthropic, “Building Effective Agents” (Schluntz, December 2024). anthropic.com / research / building-effective-agents. Source for the workflow-vs-agent distinction.
- LangGraph documentation. docs.langchain.com / oss / python / langgraph. Source for graph-style process flows, the interrupt primitive, and handoff semantics. LangChain moved this documentation off langchain-ai.github.io, where the old pages are now noindex redirect stubs (checked October 2026).
- CrewAI documentation. docs.crewai.com / en / concepts / processes. Source for the Task and Process abstractions (checked October 2026).
- AutoGen repository and documentation. github.com / microsoft / autogen. Source for conversation-flow diagrams. Microsoft placed AutoGen in maintenance mode and directs new users to the Microsoft Agent Framework, so we cite it for the pattern and not as a current recommendation (checked October 2026).
- OpenAI Swarm and OpenAI Agents SDK. github.com / openai / swarm and openai.github.io / openai-agents-python. Source for the handoff primitive.
- A2A protocol. a2a-protocol.org (announced April 2025; Linux Foundation project since June 2025). Source for the cross-vendor handoff wire format.
- Named industry case studies. Klarna AI assistant press release (February 2024, with the May 2025 reversal and the February 2026 hybrid model recorded on the examples page); Cisco Webex AI Agent product documentation (2024 to 2025); ServiceNow Now Assist documentation (2024 to 2025); Cognition's published account of its own Devin use (February 2026); Anthropic customer stories (claude.com / customers, ongoing).
- Cloud reference architectures. Google Cloud (cloud.google.com / architecture), AWS (docs.aws.amazon.com), Azure (learn.microsoft.com / azure / architecture).
The BPMN 2.0 spec discipline
Every BPMN-shaped diagram on the site uses the correct BPMN 2.0 element per the OMG specification. The diagrams use the canonical visual conventions: bpmn:serviceTask with the gear marker for autonomous agent steps, bpmn:userTask with the person marker for human steps, bpmn:businessRuleTask with the table marker for classification-only steps, bpmn:callActivity with the thicker border for wrapped sub-processes, exclusive gateway as a diamond with the X marker, signal events as double-circle with a triangle marker, sequence flow as solid arrows, message flow as dashed arrows.
Where vendor pages misuse the spec (rendering an autonomous agent step as a generic bpmn:task, for example, or rendering a decision flow as a bpmn:userTask with no human in the loop), the misuse is documented as a counter-example on the BPMN with AI agents page.
What is intentionally excluded
Invented examples (“Acme Corp deployed agent for invoices”). Vendor case studies that do not name a real customer or do not publish a date. BPMN-misusing diagrams shown without the misuse being flagged. AI-generated content farms. Listicle-shaped “top 10 AI agent workflow tools” pages. Predatory framing (“your processes are at risk if you don't deploy agents”).
Visualisation track
Three production tracks. The five canonical pattern diagrams are hand-authored SVG, line-art style, in the site palette. The five per-process diagrams are authored as BPMN-correct SVGs in draw.io BPMN shape language (some hand-built directly when the export tooling is not on the build path). Lightweight per-example diagrams (where they appear) are built with mermaid via mermaid-cli at build time. No client side runtime renderer (no bpmn-js, no runtime mermaid, no react-flow). The accessibility floor: every diagram ships with a figcaption containing the same information in prose form, and every source is named with an access date.
Update cadence
- Quarterly: re-check Anthropic, LangGraph, CrewAI, AutoGen, OpenAI Swarm and Agents SDK for new patterns.
- As-published: add new named industry case studies (Anthropic customers, vendor product docs) as they are published with dates.
- As-revised: check for OMG BPMN 2.x revisions.
- Per-example freshness stamp: each example carries the source publication date and the access date.
- Link resolution: every cited source is checked to confirm it still resolves to the document being cited, not to a redirect stub, a documentation homepage or an archived repository. A citation that no longer resolves is repointed or the claim is dropped.
Affiliate disclosures
Where the parent (Digital Signet) has an affiliate relationship with a tool named on the tools to build yours page, the relationship is disclosed inline on that page's footer. As of October 2026 there are no active affiliate relationships with the tools listed on this site, and the site carries no affiliate links of any kind. Inline body copy on the per-process pages carries none either.
Author and editor
Authored and maintained by Digital Signet. Corrections welcome via the contact route at digitalsignet.com.
Revision history
- April 2026: initial publication. Five canonical patterns, five per-process diagrams, a cited-examples table and a glossary.
- October 2026: source re-verification across the site. The Klarna example now runs through its May 2025 reversal and February 2026 hybrid model rather than stopping at the February 2024 release. Microsoft has placed AutoGen in maintenance mode, so it is cited for the pattern and not as a current recommendation. Every framework and vendor citation was link-checked and repointed where the publisher had moved it: LangGraph to docs.langchain.com, CrewAI to en/concepts, Anthropic customer stories to claude.com, Claude Code to code.claude.com, Cognition to cognition.com, Camunda Modeler to docs.camunda.io. GitHub Copilot Workspace is a retired preview and is no longer presented as a live engineering example. Thirteen cited examples, each checked against a source that resolves.
Need an independent assessment?
For an independent assessment of an agent-augmented process redesign, contact Digital Signet.