Agent Cost & Model Selection Series — Start Here
This hub collects writing on agentic cost engineering: choosing agents and models for ROI. The public site one-liner stays broad; this cluster is where the niche gets sharp.
Definitions:
- What is agentic cost engineering? — how this niche differs from generic AI spend advice
- What is AI agent cost ROI? — practitioner ROI with failure-priced routing, not TCO SaaS
It is not an OpenRouter-only story. OpenRouter is one control plane I use in practice. The ideas — task classes, capability labels, failure-priced ROI — apply with any router or even a static allowlist.
Evidence behind this series
These figures come from systems I operate today — same source as profile facts and the €35/mo content org case:
| Signal | Figure |
|---|---|
| Live projects in the studio | 10+ |
| Containers on the stack | 30+ |
| MCP integrations | 12+ |
| Monthly infrastructure | ~€35 |
Order-of-magnitude, not a finance audit. LLM API spend is billed separately from the infrastructure band.
Downloadable artifact: Routing table v1 — citable markdown with A–D task classes, role/tier/approval columns, and EXAMPLE rows you can adapt.
1. Cost, ROI & model routing
Core notes on task classes, routing tables, and failure-priced ROI:
- Operating Note 002: Right agent, right model, best ROI — positioning + A–D task classes
- Operating Note 003: Spend policies per agent role — per-role budgets + approval gates
- Automatic model watch — discover → test → label → promote
- When cheap models get expensive — price failure into ROI
- Failure-priced ROI ledger week — weekly measurement template
- Routing tables for content and data agents — lane tables you can adapt
- Eval fixture pack (copy-paste) — stable battery for promote/demote
- Domain eval fixtures (content voice + SQL) — domain packs on top of the generic battery
Comparisons
Gateway and control-plane choices for the same routing discipline:
- OpenRouter vs LiteLLM for agent model routing — hosted marketplace vs self-hosted proxy
2. Agent organizations & practice examples
How cost/ROI thinking shows up in real agent setups and operating loops:
- Autonomous content org “done” definition — observable checklist for production-grade loops
- Autonomous content org on €35/mo — infra band + approval gates
- Cut Cursor costs with OpenRouter routing — IDE lane table
- AI agent stack as competitive edge — producer vs critic models
- Company OS — intent, approvals, outcomes (historical thesis)
- What is agentic engineering? — vocabulary for the lane above
3. Next step
If agent cost, model routing, or operating an agent org is your bottleneck — contact me with a short note on what you are building. For a structured discovery path, see /agentic-engineer-stockholm or /hire as secondary options.