What tools actually make up an agentic platform
A neutral inventory, not a leaderboard. Six families, each mapped to the architecture plane it serves — because the pattern is vendor-neutral even when your stack isn't.
The short answer
You need four things: a contract layer (Score), a policy gate (OPA), a builder lane (any competent agent), and telemetry (OpenTelemetry). Everything else is ergonomics. The builder lane is the easy part — the platform around it is the discipline.
Workload specification
serves the contract layerScore
score.devOpen workload specification — declare what to deploy; the platform decides how. The demo's contract layer.
Kubernetes manifests / Helm
kubernetes.ioThe raw substrate Score and golden paths render into. What agents must NOT hand-write unguarded.
Policy engines
serves governanceOPA / Rego
openpolicyagent.orgThe de-facto policy-as-code engine. Validates before mutation; fails closed. Powers the demo's 42-check gate.
Kyverno
kyverno.ioKubernetes-native policy as manifests. Same plane, k8s-first ergonomics.
Agent lanes (the builders)
serves reasoningGitHub Copilot agents
github.com/features/copilotCoding agent lane tied to the GitHub/Azure ecosystem. Strong builder; the platform supplies the guardrails.
Claude Code
anthropic.com/claude-codeHeadless-capable CLI agent lane. A proven adversarial reviewer in cross-model setups.
Codex CLI
openai.com/codexOpenAI's terminal agent lane. Fills the same builder slot; slots should be swappable by design.
GLM-class models
z.aiCost-efficient reasoning family. The pattern is vendor-neutral: any competent lane works inside the guardrails.
Agent plumbing
serves tool accessMCP (Model Context Protocol)
modelcontextprotocol.ioOpen protocol connecting models to tools and data. Standard way an agent's tool surface is exposed.
IDP portals
serves the platform surfaceBackstage
backstage.ioOpen-source developer portal; the IDP-era reference. Golden paths here become agent contracts in the agentic era.
Port
getport.ioCommercial internal developer portal. Frequently appears in "agentic engineering platform" conversations.
Humanitec
humanitec.comPlatform orchestration — scoring workloads against golden paths. Same contract idea, commercial packaging.
Observability
serves traces & evalsOpenTelemetry
opentelemetry.ioThe telemetry standard. Its GenAI semantic conventions are the emerging vocabulary for agent traces.
Maintenance note: this page is dated and reviewed quarterly — a landscape in a category this fast goes stale quietly. The reference stack in the demo is deliberately all-free, all-open: Score, OPA, Kubernetes rendering, MCP. See these tools working together in the architecture.
Quick answers
What tools do you need for agentic platform engineering?
Four families, mapped to the architecture: a workload specification for contracts (Score), a policy engine for governance (OPA or Kyverno), one or more agent lanes for reasoning (Copilot, Claude Code, Codex, GLM-class — swappable by design), and telemetry for traces and evals (OpenTelemetry). The lanes are the easy part; the platform around them is the discipline.
Is this an endorsement of any of these tools?
No. It is a dated inventory (October 2026, reviewed quarterly) of what appears in production stacks and discussions, mapped to the plane each serves. The demo happens to be built on Score, OPA, and MCP because they are open and free to run — the pattern is deliberately vendor-neutral.
Ready for the leap?
Partner with Adventure On The Wave to build governed, agentic platform capability — architecture, guardrails, and the human authority model to match.
A strategic initiative by Adventure On The Wave