Platform Archetypes Overview
🧭 Platform Archetypes — DevOps & SRE AI Platforms (2025 Atlas)
Understand the major categories of AI platforms and when to use them. This file condenses the Atlas “Platform Archetypes Overview” into a standalone guide.
Last updated: 2025-10-11
Why archetypes?
No single product spans triage → RCA → remediation → post‑incident learning at a “best‑in‑class” level. Teams succeed by composing 2–3 archetypes that complement each other. Use this guide to choose the brain (triage/RCA), the hands (provisioning/change), and the orchestrator (workflows/approvals/audit).
🔎 Quick comparison matrix
| Archetype | Best fit | Representative platforms | Core strength | Typical limits | Typical U‑model highs |
|---|---|---|---|---|---|
| 🔭 Observability‑First | Rapid triage, RCA, incident comms | Dynatrace Davis AI; Cisco (Splunk) AI Agents; Datadog Bits AI & Agents; Elastic AI Assistant; New Relic AI | Multi‑signal correlation; causal/hypothesis reasoning; live dashboards; post‑mortems | Usually no direct apply beyond playbooks; change control via integrations | UDM, UOM (high); UAM/UEOM (med‑high); UPM (low) |
| ⚙️ Provisioning‑Focused | Safe infra/app changes with guardrails | Azure Copilot (Agent Mode); DuploCloud; Qovery; Kuberns | IaC diff/plan → approve → apply → verify → rollback | Lighter AIOps; limited evidence for RCA; relies on external obs | UPM (high); UAM (med); UOM/UDM (low‑med) |
| 👨💻 Developer‑Centric & Frameworks | Code/PR changes; build bespoke agents | AWS Strands SDK; Atlassian Rovo Dev; GitHub Copilot Coding Agent; Zencoder; JFrog Fly; Azure AI Agent Service (Foundry) | Planning/orchestration, CI fixes, AgentOps (traces/evals) | Not a runtime ops console; limited direct infra apply | UAM (med‑high); UKM/UEOM (varies); UPM (low‑med) |
| 🏢 Enterprise Orchestrators | Cross‑team workflows, approvals, CMDB | ServiceNow Agent Orchestrator; Salesforce Agentforce (OpsAI); PagerDuty AIOps | Ticket/change graph; HIL approvals; runbooks; comms | Deep telemetry depends on obs stack; limited native apply | UAM/UDM (high); UKM/UEOM (med‑high); UPM (med) |
| 📊 Data & MLOps | AI quality, lineage, pipeline ops | Databricks Agent Bricks; Snowflake Cortex Agents; Dataiku AI Agents | Evals/guardrails, lineage, cost/quality governance | Infra AIOps out of scope; limited live ops | UKM/UEOM/UAM (med‑high); UOM/UDM/UPM (low‑med) |
| 🎯 Specialized Domain | Deep accuracy for a niche | IBM AskIAM (IAM); Solo.io Kagent (K8s) | Precision in a narrow domain; strong playbooks | Intentional breadth limits | High in one model (e.g., UPM for IAM; UDM/UOM for K8s) |
Reading “U‑model highs” — which baseline models a given archetype tends to score highest on: UOM (Observability), UDM (Diagnostics), UAM (Activities), UPM (Provisioning), UEOM (Ontology), UKM (Knowledge).
Decision helper — pick by job‑to‑be‑done
- “We need faster triage with evidence.” → Start 🔭 Observability‑First; add 🏢 Orchestrator for comms/approvals.
- “We want safe, auditable changes.” → Start ⚙️ Provisioning‑Focused; add 🔭 Observability‑First for verify & SLO checks.
- “We’ll build custom agents with governance.” → Start 👨💻 Developer‑Centric & Frameworks; add 🏢 Orchestrator for enterprise routing.
- “We live inside tickets/CMDB across many teams.” → Start 🏢 Orchestrator; integrate 🔭 Observability‑First and ⚙️ Provisioning‑Focused lanes.
- “We must measure AI quality & data lineage.” → Start 📊 Data & MLOps; integrate with your agent runtime.
- “We need accuracy in a narrow domain (IAM/K8s).” → Add a 🎯 Specialized agent alongside your main brain/hands.
Evidence to request per archetype
Ask vendors to provide exportable artifacts (permalinks/JSON/Markdown) so you can reproduce claims.
- 🔭 Observability‑First: Alert rule(s), grouped incident record, causal/hypothesis graph, cross‑signal drill‑down links, post‑mortem export.
- ⚙️ Provisioning‑Focused: Plan diff, approval trail, execution logs, verification read‑backs, rollback evidence.
- 👨💻 Developer‑Centric & Frameworks: Agent traces (OTel), evaluation suites/metrics, tool registry, reproducible runs.
- 🏢 Orchestrators: Workflow definitions, CMDB/graph joins (OTel→CMDB), governance tests (RBAC, tenancy), audit exports.
- 📊 Data & MLOps: Eval datasets, guardrail policies, lineage graph, agent run telemetry.
- 🎯 Specialized: Domain‑specific test corpus + success criteria; safety/guardrail proofs.
Detailed archetype notes
🔭 Observability‑First Agents
Definition. Born from APM/monitoring/logging. Excel at anomaly detection, correlation, and causal/hypothesis‑driven RCA with cross‑signal drill‑downs.
Strengths. Fast TTFC/TTRC; entity pages; SLO burn; incident narratives & post‑mortems.
Limits. Usually no direct apply beyond playbooks; change gates via external systems.
Examples. Dynatrace Davis AI; Cisco (Splunk) AI Agents; Datadog Bits AI & Agents; Elastic AI Assistant; New Relic AI.
Typical U‑profile. UDM/UOM high; UAM/UEOM med‑high; UPM low.
⚙️ Provisioning‑Focused Agents
Definition. IaC‑driven control planes for plan→approve→apply→verify→rollback under guardrails.
Strengths. Deterministic plans, policy gates, drift checks, rollback; direct cloud/K8s changes.
Limits. Limited RCA/correlation; relies on your observability for evidence.
Examples. Azure Copilot (Agent Mode); DuploCloud; Qovery; Kuberns.
Typical U‑profile. UPM high; UAM med; others low‑med.
👨💻 Developer‑Centric & Frameworks
Definition. Code/PR agents and agent frameworks with strong AgentOps (traces/evals/safety).
Strengths. Planning/orchestration; CI fixes; reproducibility; SDK & tool integration (MCP/OpenAPI).
Limits. Not a production ops console; changes often land as PRs rather than live apply.
Examples. AWS Strands SDK; Atlassian Rovo Dev; GitHub Copilot Coding Agent; Zencoder; JFrog Fly; Azure AI Agent Service (Foundry).
Typical U‑profile. UAM med‑high; UKM/UEOM variable; UPM low‑med (PR‑first).
🏢 Enterprise Workflow Orchestrators
Definition. Low‑code multi‑agent coordinators across ITSM/ITOM with CMDB and approvals.
Strengths. Incident workflows; comms; RBAC/tenancy; OTel→CMDB mappings; runbooks.
Limits. Deep telemetry via integrations; limited direct apply.
Examples. ServiceNow AI Agent Orchestrator; Salesforce Agentforce (OpsAI); PagerDuty AIOps.
Typical U‑profile. UAM/UDM high; UKM/UEOM med‑high; UPM med (via actions).
📊 Data & MLOps Agents
Definition. Agentic governance for data/ML: evals, guardrails, lineage, cost/quality.
Strengths. Evaluation harnesses; experiment tracking; trace observability for agents.
Limits. Infra AIOps out of scope; provisioning limited to platform tasks.
Examples. Databricks Agent Bricks; Snowflake Cortex Agents; Dataiku AI Agents.
Typical U‑profile. UKM/UEOM/UAM med‑high; others low‑med.
🎯 Specialized Domain Agents
Definition. High accuracy within a narrow scope (e.g., IAM or Kubernetes).
Strengths. Deep playbooks; domain primitives; safer auto‑actions in‑scope.
Limits. Limited breadth by design; must be composed with other archetypes.
Examples. IBM AskIAM (IAM); Solo.io Kagent (K8s).
Typical U‑profile. One or two models high (e.g., UPM in IAM; UDM/UOM in K8s).
Integration patterns (compose your stack)
1) Observability brain + Orchestrator + Provisioning hands
- Use when: You want fast RCA with evidence, human‑in‑the‑loop approvals, and safe changes.
- Shape: 🔭 + 🏢 + ⚙️
2) Framework + Orchestrator (enterprise guardrails)
- Use when: You’ll build tailored agents and need governance and routing.
- Shape: 👨💻 + 🏢 (+ 🔭 for evidence)
3) Specialist + Observability brain (targeted wins)
- Use when: K8s/IAM pain dominates; add focused automation next to generic triage.
- Shape: 🎯 + 🔭 (+ ⚙️ if changes are frequent)
Evaluation checklist (per archetype)
- Evidence exportability: permalinks/JSON/Markdown to reproduce claims.
- Cross‑signal drill‑down: mandatory for 🔭 at Level ≥ 3 (see UOM).
- Change safety: diffs, approvals, rollback proof for ⚙️ at Level ≥ 3 (see UPM).
- AgentOps: traces, evals, safety guards for 👨💻 and 🏢.
- Ontology: typed entities & joins for 🏢/🔭 at Level ≥ 3 (see UEOM).
- Governance: RBAC/PII/tenancy tests for any archetype deployed at scale.
Recap — quick list
- 🔭 Observability‑First: Dynatrace Davis AI; Cisco (Splunk) AI Agents; Datadog Bits AI & Agents; Elastic AI Assistant for Observability; New Relic AI.
- ⚙️ Provisioning‑Focused: DuploCloud AI Help Desk; Qovery AI Migration Agent; Kuberns Platform; Microsoft Azure Copilot (Agent Mode).
- 👨💻 Developer‑Centric & Frameworks: AWS Strands SDK; Atlassian Rovo Dev; GitHub Copilot Coding Agent; Zencoder; JFrog Fly (agentic repo); Microsoft Azure AI Agent Service (Foundry).
- 🏢 Enterprise Workflow Orchestrators: ServiceNow AI Agent Orchestrator; Salesforce Agentforce (OpsAI); PagerDuty AIOps.
- 📊 Data & MLOps Agents: Databricks Agent Bricks; Snowflake Cortex Agents; Dataiku AI Agents.
- 🎯 Specialized Domain: IBM AskIAM (IAM); Solo.io Kagent (K8s).
Notes
- Scoring shorthand (Y/H, Y/M, P/M…) maps to 0–4 levels per the baseline models.
- Use the same incident corpus and change sets across vendors for fair comparisons.