Linton proposes tiered model agent execution

Over-provisioning coding agents with top-tier reasoning models for routine execution tasks is a costly anti-pattern. Standardizing on a tiered orchestration loop that dynamically routes sub-tasks to cheaper models or lower effort levels is key to scaling agentic development.

• –Avoid Over-provisioning: Using max-effort frontier models for simple, boilerplate code execution wastes significant compute and budget.
• –Variable Effort Levels: Dynamic routing allows simple tasks to run on faster, cheaper models, reserving heavy reasoning models exclusively for complex edge cases.
• –Loop Segmentation: Separating orchestration (planning), execution (writing), and review (linting, testing, verifying) creates distinct checkpoints that prevent error propagation.
• –Developer Impact: Teams adopting tiered agent workflows will see a dramatic drop in token consumption and execution latency without sacrificing code quality.