Cost-Per-Pass, Not Cost-Per-Call#

Practitioners price LLMs by the per-token rate on the provider’s pricing page. For autonomous agents, that number is misleading. Two layers of indirection sit between the per-token rate and the cost you actually pay to get work done: variable prompt sizes turn per-token into per-call, and variable pass rates turn per-call into per-pass. Each layer can invert the ranking.

For autonomous fleets where failed attempts trigger reviewer cycles, retries, and reputational drag, cost-per-pass is the only metric that ranks models correctly. This article shows how to compute it, when it dominates, and where the cheapest-per-token model becomes the most expensive in production.

CI/CD Cost Optimization#

CI/CD costs grow quietly. A team of ten pushing five times a day, running a 15-minute pipeline on 4-core runners, burns through 2,500 build minutes per week. On GitHub Actions at $0.008/minute for Linux runners, that is $20/week. Scale to fifty developers with integration tests, matrix builds, and nightly jobs, and you are looking at $500-$2,000/month before anyone notices.

The fix is not running fewer tests or skipping builds. It is eliminating waste: jobs that use more compute than they need, caches that are never restored, full builds triggered by README changes, and runners sitting idle between jobs.

Tiered-LLM Tooling: Local by Default, Escalate to Frontier#

When you build a chat or ops interface backed by an LLM, paying a frontier model for every interaction is wasteful — most interactions are cheap lookups, summaries, and routing. A tiered design serves the high-frequency majority with a small local model (e.g. an Ollama-served model on a GPU you already have) and escalates to a frontier model (e.g. Claude) only for the hard minority.

An agent fleet wired to a high-volume trigger source — channel mentions, queue events, webhooks — pays full cost on every cycle, even when the trigger is noise. A classifier placed in front of the main agent decides which triggers deserve a real cycle and which to drop. The pattern is old; what is new is that local LLMs make the classifier cost effectively zero, which flips the arithmetic in the pattern’s favor for cases that previously didn’t justify the latency.

The Cost Optimization Hierarchy#

Cloud cost optimization follows a hierarchy of impact. Work from the top down – fixing the wrong tier of commitment discount matters far less than shutting down resources nobody uses.

1. Eliminate waste – turn off unused resources, delete orphaned storage
2. Right-size – match instance sizes to actual usage
3. Use commitment discounts – reserved instances, savings plans, CUDs
4. Shift to spot/preemptible – for fault-tolerant workloads
5. Optimize storage and network – tiering, transfer patterns, caching
6. Architect for cost – serverless, auto-scaling, multi-region strategy

Eliminating Waste#

The fastest cost reduction comes from finding resources that serve no purpose. Every cloud provider accumulates these: instances left running after a test, snapshots from decommissioned servers, load balancers with no backends, unattached disks.

What Capacity Planning Solves#

Running out of capacity during a traffic spike causes outages. Over-provisioning wastes money continuously. Capacity planning is the process of measuring what you use now, projecting what you will need, and ensuring resources are available before demand arrives. Without it, you are either constantly firefighting resource exhaustion or explaining to finance why your cloud bill doubled.

Capacity planning is not a one-time exercise. It is a recurring process – monthly for fast-growing services, quarterly for stable ones.

Kubernetes Cost Audit and Reduction#

Kubernetes clusters accumulate cost waste silently. Resource requests padded “just in case” during initial deployment never get revisited. Load balancers created for debugging stay running. PVCs from deleted applications persist. Over six months, a cluster originally running at $5,000/month can drift to $12,000 with no corresponding increase in actual workload.

This operational plan works through cost reduction systematically, starting with visibility (you cannot cut what you cannot see), moving through quick wins, then tackling the larger structural optimizations that require data collection and careful rollout.

Kubernetes FinOps: Decision Framework for Cost Optimization#

FinOps in Kubernetes is the practice of bringing financial accountability to infrastructure spending. The challenge is not a lack of cost-saving techniques – it is knowing which ones to apply first, which combinations work together, and which ones introduce risk that outweighs the savings. This article provides a structured decision framework for selecting and prioritizing Kubernetes cost optimization strategies.

The Five Optimization Levers#

Every Kubernetes cost optimization effort works across five levers. Each has a different risk profile, implementation effort, and savings ceiling.

Multi-Cloud vs Single-Cloud Strategy#

Multi-cloud is one of the most oversold strategies in infrastructure. Vendors, consultants, and conference speakers promote it as the default approach, but the reality is that most organizations are better served by a single cloud provider used well. This framework helps you determine whether multi-cloud is actually worth the cost for your situation.

The Default Answer Is Single-Cloud#

Start with single-cloud unless you have a specific, concrete reason to go multi-cloud. Here is why.

Kubernetes Cost Optimization#

Most Kubernetes clusters run at 15-30% actual CPU utilization but are billed for the full provisioned capacity. The gap between what you reserve and what you use is pure waste. This article covers the practical workflow for finding and eliminating that waste.

The Cost Problem: Requests vs Actual Usage#

Kubernetes resource requests are the foundation of cost. When a pod requests 4 CPUs, the scheduler reserves 4 CPUs on a node regardless of whether the pod ever uses more than 0.1 CPU. The node is sized (and billed) based on what is reserved, not what is consumed.