chart/AgenticOrgChart.com
menu
Patterns
Reference
Examples gallery / 2026

AI Agent Org Chart Examples: Cited Public Deployments (2026)

Every example below cites a real public source. Where a source does not name a real customer or deployment with a publication date, the example is not on this page.

The discipline behind this page: every diagram carries the source URL, the publication date, and the access date in the figcaption. Examples are grouped by topology pattern. Each example links to the relevant pattern page on this site, the source URL, and (where applicable) the by-industry sub-page.

Single-agent topology

Examples drawn from coding agents and assistant-style deployments.

Most production agent deployments are single-agent. The two cited examples below illustrate the shape: one autonomous (Devin), one assistant-style (claude-code). Read more on the single-agent topology page.

Anthropic claude-code single-agent engineering loopAn engineer supplies a coding goal. A single coding agent reads the codebase, edits files, runs tests, and iterates until tests pass or the agent yields back to the engineer.Engineergoal inCoding agentread / edit / testread filesedit filesrun testsshell execloop until tests pass
Cognition Devin and Anthropic claude-code: single-agent engineering loop. One agent reads, edits, runs tests, and iterates.Sources: Cognition AI, “Introducing Devin” (12 March 2024) cognition.ai/blog/introducing-devin; Anthropic claude-code product page anthropic.com/claude-code. Accessed 30 April 2026. Related: /single-agent/, /by-industry/engineering/.
Multi-agent peer topology

CrewAI and AutoGen patterns from official documentation.

CrewAI three-role crew topologyA CrewAI crew composed of a researcher agent, a writer agent, and a reviewer agent collaborates on a goal. Each agent has a defined role and toolset; tasks pass sequentially through the crew.Operatorgoal inResearcherweb search, RAGWriterdrafts contentReviewerchecks against criteriaFinal artefactdelivered outputresearch → draftdraft → review
CrewAI three-role crew (researcher + writer + reviewer). The pattern in the official CrewAI examples repository.Source: CrewAI examples repo github.com/crewAIInc/crewAI-examples; CrewAI documentation docs.crewai.com/concepts/agents. Accessed 30 April 2026. Related: /multi-agent/.
AutoGen group-chat topologyAutoGen's group-chat topology: multiple agents communicate through a GroupChatManager that selects the next speaker, with shared message history visible to all.GroupChatManagerselects next speakerUserProxyCoder agentCritic agentExecutorshared message history
AutoGen group-chat with GroupChatManager selecting the next speaker.Source: AutoGen documentation, teams tutorial microsoft.github.io/autogen/stable/user-guide/agentchat-user-guide/tutorial/teams. Accessed 30 April 2026. Related: /multi-agent/.
Supervisor pattern

LangGraph’s supervisor primitive.

LangGraph supervisor with handoffsLangGraph's supervisor primitive: a supervisor node routes messages to one of several worker subgraphs by emitting a handoff. Workers may hand back to supervisor or terminate.Supervisor nodehandoff(next)Researchersubgraphtools: searchCodersubgraphtools: REPLReviewersubgraphtools: lint, evalEND (terminate)
LangGraph supervisor with handoffs.Source: LangGraph supervisor tutorial langchain-ai.github.io/langgraph/tutorials/multi_agent/agent_supervisor. Accessed 30 April 2026. Related: /supervisor-pattern/.
Hierarchical agents

LangGraph hierarchical-teams.

LangGraph hierarchical-teams topologyLangGraph's hierarchical-teams primitive: a top-level supervisor delegates to two team-level supervisors, each of which manages two worker agents. Each layer is a subgraph in its own right.Top supervisorroutes to teamsResearch team leadsubgraphWriting team leadsubgraphSearchwebSearchdocsDrafterv1Polisherv2
LangGraph hierarchical-teams. Top supervisor delegates to two team-lead supervisors; each lead manages two workers.Source: LangGraph hierarchical-teams tutorial langchain-ai.github.io/langgraph/tutorials/multi_agent/hierarchical_agent_teams. Accessed 30 April 2026. Related: /hierarchical-agents/.
Human-in-the-loop

Reviewer (LangGraph interrupt) and arbiter (Klarna).

LangGraph interrupt human-approval topologyLangGraph's interrupt() primitive: a graph pauses before a write-action node, surfaces a human-approval prompt, and resumes only when the approver responds.Plan nodebuilds proposalinterrupt()graph pausesApproverapprove / rejectWrite actionsystem of recordrequestreject → retryon approve
LangGraph interrupt() reviewer pattern.Source: LangGraph human-in-the-loop concepts langchain-ai.github.io/langgraph/concepts/human_in_the_loop. Accessed 30 April 2026. Related: /human-in-the-loop/.
Klarna customer-service tier-1 deflection topologyA customer message enters a tier-1 customer service agent. The agent answers FAQs and processes returns directly. Ambiguous or sensitive cases escalate to a human agent.Customerchat inTier-1 agentFAQ + returns + refundsHuman agentcomplex / sensitiveResolutionresponse / actionescalateresolve
Klarna AI assistant: tier-1 deflection plus escalation. Klarna’s February 2024 report stated the assistant handled two-thirds of customer-service chats in its first month.Source: Klarna press release (27 February 2024) klarna.com. Accessed 30 April 2026. Related: /human-in-the-loop/, /by-industry/customer-service/.
Clinical AI agent with mandatory clinician approvalA clinical agent drafts a recommendation from patient data. A clinician must review and approve before the recommendation is added to the medical record.Patient dataEHR / labs / imagingClinical agentdrafts recommendationflags evidenceClinicianreviews evidenceapproves / rewritesMedical record (after approval)on approve
Clinical AI agent with mandatory clinician approval. Pattern observed across Mayo Clinic Platform and academic clinical-AI deployments.Source: Mayo Clinic Platform mayoclinicplatform.org; FDA AI/ML SaMD guidance fda.gov/medical-devices/software-medical-device-samd. Accessed 30 April 2026. Related: /by-industry/healthcare/.
JPMorgan COIN document-extraction topologyA document agent extracts terms from contract documents. Material clauses are flagged and routed to a human reviewer; routine terms are written to a structured store.Contractscanned PDFExtraction agentparses clausesflags materialCounsel reviewermaterial clausesStructured storeroutine termsflagauto-write
JPMorgan COIN: contract-extraction agent with counsel review. Document parsing autonomously, counsel sign-off on flagged material clauses.Source: JPMorgan public communications and Bloomberg (28 February 2017) coverage of the Contract Intelligence platform; FCA AI in Financial Services Survey (October 2024) fca.org.uk. Accessed 30 April 2026. Related: /by-industry/financial-services/.
Evaluator-optimiser pattern

Self-refine (single-agent precursor) and the SWE-Bench coding-agent loop.

Self-refine single-agent loopA single agent generates an output, critiques it, and refines based on its own critique. The same model plays both roles in alternating turns.Single agentgenerate / critique / refinesame model, three rolesself-loop until quality criteria metTaskRefined output
Self-refine single-agent loop. The single-agent precursor to evaluator-optimiser.Source: Madaan et al., “Self-Refine: Iterative Refinement with Self-Feedback” (March 2023) arxiv.org/abs/2303.17651; related: Shinn et al., “Reflexion” (March 2023) arxiv.org/abs/2303.11366. Accessed 30 April 2026. Related: /evaluator-optimiser/.
Supervisor + escalation (IT support)

Cisco TAC AI agent.

Cisco TAC AI agent topologyA customer logs a support case. A triage agent classifies the case and either resolves it from documentation or routes it to a senior engineer for complex network issues.Customersupport caseTriage agentclassifies severityretrieves docsDoc-resolution agentknown issuesSenior engineercomplex / sev1Diagnostic agentruns network probesknowncomplexdiagnose
Cisco TAC AI agent. A triage agent classifies severity, resolves known issues from documentation, escalates complex cases to senior engineers, and delegates network probing to a diagnostic agent.Source: Cisco AI in customer experience disclosures and the Cisco Live 2024 keynote (June 2024). Pattern shape consistent with the published Cisco AI Solutions Engine architecture overview at cisco.com/c/en/us/solutions/artificial-intelligence. Accessed 30 April 2026. Related: /supervisor-pattern/, /human-in-the-loop/.

What is intentionally excluded

Invented “Acme Corp deployed five agents” examples. Vendor case studies that do not name a real customer or do not publish a date. AI-generated content farms. The discipline is documented on the methodology page.

Where a useful published deployment exists but the agent topology is not publicly disclosed, the example is not represented on this page. Corrections welcome via the contact note in the methodology page.