cleandataengineering.aiAgent Harness, drawn simply
You are here: Architecture › On Azure › What stays ours

Agent Harness Tutorial · Architecture

What Azure gives, what we build, what each agent does

For each concern, how much does Azure give us, how much do we code once for every agent, and how much does each agent team still write? Azure gives us the plumbing. The parts that make an agent trustworthy stay in our code.

AZURE GIVES (x)OURS, CODED ONCE (y)EACH AGENT (z)L1 ContextL2 Model callL4 LoopL5 State18 KnowledgeL3 ToolsL6 ContractsL8 Queue17 Hand-offChannelsL7 PolicyL10 FailuresL14 ApprovalsL16 IdentityL9 ReplayL11 ObservabilityL13 CostL12 EvalL15 Release
Average: platform 32% · ours 47% · each agent 21%. The rule: z shrinks over time; code two teams write moves into y, and when Azure offers what we built, y moves into x.

What stays ours, in eight tiles: the control loop, context assembly, policy positions, tool design, the replay recorder, the span contract, undo and idempotency, and cost per outcome.

For every concern's choice in detail, with the Azure and open-source options next to it, see Tech stack.

Want every detail? The long version is in the Series: what Azure gives, what we build once.