Much of the work performed by AI coding agents consists of I/O operations, such as reading and generating files, rather than inference. Using high-performance models for these tasks is a major factor in driving up token costs.
Pricing & LimitsSpotifyPortalAiKA ModesClaude Code
Spotify Reduces Claude Code Token Consumption by 90% Using AiKA Modes via Portal
This article is a translation. Read the Japanese original
Developer token costs are on an upward trend; some engineering leaders report paying between $200 and $500 per developer per month, with some cases exceeding $2,000.
In contrast, the approach utilizing AiKA Modes provided by Portal delegates routine tasks to cheaper models. A Mode is a declarative agent where users can specify instructions, models, and MCP tools, requiring no infrastructure or API key management.
Specific examples include "bulk-reader," which uses Gemini 2.5 Flash when reading large volumes of files, and "code-writer," which generates code boilerplate. By utilizing these, it becomes possible to reserve high-performance models only for scenarios requiring advanced inference.
To automate this workflow, a Claude Code plugin called "shunt" was developed. This plugin automatically instructs a switch to Portal's "bulk-reader" when a file size exceeds a certain threshold.
Source: Portal by Spotify cut my Claude Code token usage by 90% (Hacker News Frontpage, 2026-09-05)