Cloud cost optimization works when spend data, utilization, and the architecture trade-off live on the same row. The agent reads AWS Cost Explorer, Datadog, and GCP Billing, drafts an optimization memo, and routes it to the engineering lead. Dock holds the proposal, reviewer, decision, and date. The agent re-fetches numbers when a memo lands for review, so the lead approves against current state.
AWS Cost Explorer, Datadog, and GCP Billing stay the system of record for the raw data. Dock is the system of record for what the agent interprets. Each Dock row carries a pointer back to the platform record, agent identity, decision, reviewer, and timestamp. The agent re-fetches platform data via fresh API reads when it needs current state.
The Optimization Memos table
| Workload | Spend signal | Utilization signal | Agent proposal | Eng-lead decision | Reviewer | Date |
|---|---|---|---|---|---|---|
| ingest-api (us-east-1) | AWS CE: $14,200/mo on m6i.4xlarge fleet | Datadog: p95 CPU 22%, mem 38% | Right-size to m6i.2xlarge, est. $6,800/mo saved | Approved, schedule for next deploy window | priya.k | 2026-05-27 |
| ml-batch (GCP) | GCP Billing: $9,400/mo on n2-standard-32 | Datadog: 6h/day idle, 18h queue depth zero | Move to Spot + autoscale to zero, est. $5,100/mo saved | Approved with rollback runbook | priya.k | 2026-05-28 |
| analytics-warehouse | AWS CE: $22,000/mo Redshift ra3.4xlarge x4 | Datadog: query concurrency peaks 2/8 slots | Hold. Concurrency headroom needed for Q3 launch | Declined, revisit August | dan.m | 2026-05-29 |
Each row links back to the Cost Explorer report, the Datadog dashboard snapshot, and the GCP billing export the agent read. The decision column is the human commitment. The reviewer column attaches a named principal, never a service account. See agent identity lifecycle for why that distinction matters.
The workflow
The agent runs weekly. It pulls the last 30 days of Cost Explorer line items, the matching Datadog utilization for each tagged workload, and GCP billing for the multi-cloud pieces. It writes one row per candidate workload with the spend signal, the utilization signal, and a proposed change. It does not change infrastructure. It drafts.
The engineering lead opens the table, reads the memo, checks the linked dashboards, and writes Approved, Declined, or Hold with a reason. Approved rows flow to a Terraform PR the agent opens in a separate repo, linking back to the Dock row. The merge reviewer sees the cost rationale and the human approval together. This is the dangerous ops contract: agent proposes, lead commits, apply is a separate human step.
Why it matters
The FinOps Foundation defines the practice as making timely, data-driven trade-offs between speed, cost, and quality through collaboration between engineering and finance (FinOps Foundation). The AWS Well-Architected Cost Optimization Pillar names expenditure awareness and optimize-over-time as core capabilities (AWS Well-Architected). Both assume the trade-off is recorded somewhere a human can defend later. Dock is that record. The agent reads three platforms, the lead decides once, and the decision sits next to the numbers that produced it. Finance reads the same table from the other side, which is the Dock for Finance view of the same workflow.
If the audit later asks why ml-batch moved to Spot, the answer is the row: the utilization snapshot, the proposed saving, the lead who approved, the date. That is what agent audit and compliance means in a cost context.
Try it
Read the cluster pillar at Dock for DevOps, then the architectural frame in Cloud 2.0 for engineering.
FAQ
Does the agent change infrastructure directly? No. It writes a memo to a Dock row and, on approval, opens a Terraform PR. The apply is a separate human step.
What stays in AWS, Datadog, and GCP? The meters, the metrics, and the bills. Dock holds the interpretation: which workload, which proposal, which reviewer, which date.
How does the lead know the numbers are current? The agent re-fetches Cost Explorer, Datadog, and GCP Billing on review. The row shows the read timestamp.
Who is on the hook for the decision? The named reviewer in the row. The agent drafted, the lead committed. The audit trail names both.