Frontier Investigation AgentsCost
GCP Rightsizing turns cloud cost analysis into safe, closed-loop action. AgenticOps is the discipline of running production cloud operations through autonomous AI agents — under team policy, with brokered credentials, sandboxed execution, deterministic data tokenization, and tamper-evident audit — so a rightsizing or waste-elimination finding from GCP Rightsizing becomes a reviewed, reversible change instead of a spreadsheet nobody actions.
CloudThinker's Frontier Investigation Agents investigate the signal, find the why and propose the fix, or run the safe action your policy allows. Your engineers approve, and the agents verify the outcome.
GCP Rightsizing continuously watches Cost Explorer, CUR/billing exports, and utilization metrics to surface concrete savings signals — underutilized EC2/RDS instances, idle and unattached resources, expiring or under-covered commitments, and spend anomalies — each tied to a specific resource ARN and a dollar/month estimate rather than a vague trend.
Each finding is enriched with context before any action: real utilization percentiles over a lookback window, tags and owner, environment (prod vs non-prod), workload criticality, and dependency/attachment checks. The agent computes the expected monthly savings, the rollback path, and a confidence score, and drops low-signal or in-use resources so only credible, reversible changes advance.
Approved changes are executed through brokered, scoped credentials in a sandboxed session — resize an instance, purchase or rebalance a Savings Plan to a target coverage, delete a confirmed-orphaned volume after snapshot, or schedule non-prod shutdown. Graduated autonomy (L1 Notify, L2 Suggest, L3 Approve, L4 Auto) lets teams keep engineers on the loop: a snapshot-then-delete of an untagged idle disk may run at L4, while a prod rightsize or a multi-year commitment stays gated at L3 approval.
After every change the agent confirms the outcome — the resource is healthy at its new size, the commitment coverage hit target, the deleted resource had a restorable snapshot — and re-measures realized spend against the estimate. Every step (who connected, what was reached, what changed, who approved) lands in a tamper-evident audit trail, and any regression triggers an automatic rollback or re-open of the finding.
$ agenticops plan --skill "GCP Rightsizing" --scope prod-account-4021
FIX PLAN (3 findings · est. savings $4,180/mo) autonomy: on-the-loop
[1] Rightsize i-0a9f...c31 (m5.4xlarge → m5.xlarge)
p95 CPU 9% / mem 22% over 30d · est. -$412/mo
rollback: resize back to m5.4xlarge (snapshot-free) → L3 APPROVE
[2] Delete idle EBS vol-07be... (unattached 41d, gp3 500GB)
no attachment · owner tag missing · est. -$40/mo
guardrail: snapshot before delete, 7d retention → L4 AUTO
[3] Increase Savings Plans coverage compute 68% → 90%
on-demand baseline stable 45d · 1yr no-upfront · -$3,728/mo
rollback: n/a (commitment) — requires human sign-off → L3 APPROVE
Apply [1] [2]? auto-applying [2] under policy. verifying...
✓ vol-07be snapshot snap-0c2 created, volume deleted, $40/mo confirmed
⧗ awaiting approval on [1] [3] (audit: run 7f3a-c domain-brokered)Analyze Compute Engine VM, Cloud SQL, Persistent Disk, and serverless (Cloud Functions/Cloud Run) utilization to identify right-sizing opportunities. Uses Cloud Monitoring metrics with anti-hallucination rules for E2 shared-core instances, sole-tenant nodes, preemptible/spot VMs, SUD eligibility, peak vs average analysis, CUD-aware savings, and estimated monthly savings calculations.
More automations in the same category.
Agents investigate every issue across application, cloud, security, database and Kubernetes, find the why and propose the fix. Your engineers approve, under your policies and audit trail.