Intro & scope
Use cases where an LLM plans or executes multi-step work across business systems (mail, ERP, CRM, browser/UI, ticketing) with human approval for risky actions. Patterns: AGENT.BACKOFFICE (internal process agents), AGENT.FRONTOFFICE (customer/market-facing), AGENT.PERSONAL (individual assistants), AGENT.PLATFORM (generic runtimes, governance).
Scope boundaries: AGENT-002 (low-code engine) and AGENT-013 (multi-agent orchestration) are generic platforms; domain agents (AGENT-004, 006, 014, 023-031) reuse them. AGENT-003/021 (computer use) are the fallback when no API exists. AGENT-025 (AgentOps) governs all others. AGENT-008 (CV screening, high risk) vs AGENT-033 (sourcing outreach, high risk): screening of received applications vs. proactive sourcing. Pure document extraction belongs to the DOC family, chat to CONV.
Use cases
| ID | Title | Pattern | Maturity | Path | HW tier | AI Act risk |
|---|---|---|---|---|---|---|
| AGENT-001 | Email inbox triage & auto-response agent | AGENT.BACKOFFICE | mainstream | B | HW2 | limited |
| AGENT-002 | LLM-powered back-office workflow automation (n8n, Power Automate & co.) | AGENT.PLATFORM | mainstream | A | HW2 | minimal |
| AGENT-003 | Computer-use / browser agent for legacy UI automation (agentic RPA) | AGENT.PLATFORM | early-adopter | A | HW3 | depends |
| AGENT-004 | Order entry automation (email/PDF orders into ERP) | AGENT.BACKOFFICE | mainstream | A | HW2 | minimal |
| AGENT-005 | Procurement agent: sourcing, quote comparison, supplier communication | AGENT.BACKOFFICE | early-adopter | A | HW3 | minimal |
| AGENT-006 | Account reconciliation & financial close agent | AGENT.BACKOFFICE | early-adopter | A | HW3 | minimal |
| AGENT-007 | Sales prospecting & outreach agent (AI SDR) | AGENT.FRONTOFFICE | early-adopter | A | HW3 | limited |
| AGENT-008 | CV screening & candidate matching | AGENT.BACKOFFICE | mainstream | A | HW2 | high |
| AGENT-009 | Employee onboarding orchestration | AGENT.BACKOFFICE | early-adopter | A | HW2 | depends |
| AGENT-010 | Travel & expense booking agent | AGENT.BACKOFFICE | early-adopter | A | HW2 | minimal |
| AGENT-011 | CRM data enrichment & hygiene agent | AGENT.FRONTOFFICE | early-adopter | A | HW2 | minimal |
| AGENT-012 | Meeting scheduling & calendar agent | AGENT.PERSONAL | mainstream | A | HW1 | minimal |
| AGENT-013 | Multi-agent systems for complex end-to-end business processes | AGENT.PLATFORM | early-adopter | B | HW3 | depends |
| AGENT-014 | End-to-end insurance claims handling automation | AGENT.BACKOFFICE | early-adopter | A | HW3 | depends |
| AGENT-015 | Customer onboarding & KYC workflow agent | AGENT.BACKOFFICE | early-adopter | A | HW3 | depends |
| AGENT-016 | IT service management: autonomous ticket resolution | AGENT.BACKOFFICE | early-adopter | A | HW3 | minimal |
| AGENT-017 | Social media management agent | AGENT.FRONTOFFICE | mainstream | A | HW2 | limited |
| AGENT-018 | Continuous compliance & controls monitoring agent | AGENT.BACKOFFICE | early-adopter | A | HW3 | minimal |
| AGENT-019 | Personal admin agent (bills, subscriptions, bureaucracy) | AGENT.PERSONAL | emerging | B | HW2 | minimal |
| AGENT-020 | E-commerce operations agent (listings, prices, customer replies) | AGENT.FRONTOFFICE | early-adopter | A | HW2 | limited |
| AGENT-021 | Data entry & system migration agent | AGENT.BACKOFFICE | early-adopter | B | HW3 | minimal |
| AGENT-022 | Agentic commerce enablement: selling through third-party AI shopping agents (ACP / UCP / AP2) | AGENT.FRONTOFFICE | emerging | A | HW0 | minimal |
| AGENT-023 | Order-to-cash agent: cash application, dunning, collections & deduction handling | AGENT.BACKOFFICE | mainstream | A | HW2 | minimal |
| AGENT-024 | Freight & shipment operations agent (carrier booking, track & trace, exception handling) | AGENT.BACKOFFICE | early-adopter | A | HW3 | minimal |
| AGENT-025 | AI agent fleet governance, observability & evaluation (AgentOps) | AGENT.PLATFORM | early-adopter | A | HW1 | depends |
| AGENT-026 | Public-administration case-processing agent (Antragsbearbeitung) | AGENT.BACKOFFICE | early-adopter | B | HW3 | high |
| AGENT-027 | Inbound RFQ-to-quote agent for make-to-order manufacturers | AGENT.FRONTOFFICE | early-adopter | A | HW3 | minimal |
| AGENT-028 | Credit analysis & credit memo preparation agent (commercial lending) | AGENT.BACKOFFICE | early-adopter | A | HW4 | depends |
| AGENT-029 | Autonomous supplier negotiation agent (tail-spend contracts) | AGENT.BACKOFFICE | early-adopter | A | HW3 | minimal |
| AGENT-030 | Automatic time capture & billing-narrative generation for professional services | AGENT.BACKOFFICE | early-adopter | A | HW2 | minimal |
| AGENT-031 | Card dispute & chargeback handling automation | AGENT.BACKOFFICE | early-adopter | A | HW2 | minimal |
| AGENT-033 | Talent sourcing & passive-candidate outreach agent | AGENT.BACKOFFICE | early-adopter | A | HW3 | high |
Shared stack
Common building blocks: agent-framework + workflow-automation (orchestration), guardrails (input/output filters, approval gates), llm-general/llm-reasoning (planning), ocr-docparse (documents), embedding-text + vector-db (memory/RAG), AgentOps tracing (AGENT-025).
| HW tier | Reference stack | Serves |
|---|---|---|
| HW0 | Hosted EU API (EU-region LLM endpoint), managed workflow tool, no local inference | AGENT-022 |
| HW1 | n8n or Temporal on a small VM (CPU only), LLM via EU API, Langfuse for traces, Postgres + pgvector | AGENT-012, 025 |
| HW2 | Single 24 GB GPU server: vLLM serving a 7-14B instruct model (e.g. Mistral/Qwen), bge-m3 embeddings, Qdrant/pgvector, Docling/OCR, n8n, Langfuse, Presidio PII filter | AGENT-001, 002, 004, 009-011, 017, 019, 020, 023, 030, 031 |
| HW3 | 1-2x 48-80 GB GPU: vLLM with a 30-70B model plus reasoning model, VLM for screenshots/PDFs, reranker, sandboxed browser (Playwright/computer-use container), LangGraph or similar multi-agent runtime, Langfuse | AGENT-003, 005-008, 013-016, 018, 021, 024, 026, 027, 029, 033 |
| HW4 | Multi-GPU node (4-8x 80 GB) or private cluster for 100B+ models with strict data residency, full audit logging | AGENT-028 |
Cross-family sharing: reuse the same vLLM endpoint, vector store and OCR service with DOC (extraction), CONV (assistants), KNOW (RAG) and SEC (guardrails); one Langfuse/AgentOps instance and one approval-gate service across all agents. High-risk cases (AGENT-008, 026, 033; credit/KYC where applicable) need human oversight, logging and Annex III conformity work regardless of tier.