Portal by Spotify Cut Claude Code Token Usage by 90%

Most of what an AI coding agent does isn't thinking. It's I/O. Reading five files to answer one question. Generating a test that follows the same pattern as the twenty next to it. Thousands of tokens, zero reasoning — fed to a frontier model wildly overqualified for the task.

Spotify's Dimitri Mazmanov published the fix: a Claude Code plugin called shunt that routes I/O-heavy work to cheaper models via Portal's AiKA Modes. The headline: 90% fewer tokens on bulk reads and boilerplate generation.

What shipped

Shunt is three layers on Portal's ephemeral agent runtime. PreToolUse hooks block reads on files over 350 lines (configurable), redirecting Claude to a /bulk-reader skill. Two bash scripts wrap Portal CLI calls — bulk-read sends files + questions to the reader mode; code-write generates boilerplate matching existing patterns, writing directly to disk. Both use Gemini 2.5 Flash ($0.15/M input tokens vs. Claude's $3/M). The hooks enforce routing; Claude can't ignore them.

Why it matters

Gartner predicts AI coding costs will surpass the average developer salary by 2028. A quarter of engineering leaders already burn $200–$500 per developer per month on tokens; some are past $2,000. Shunt's approach is notable because it's not a new platform — it's a configuration: two mode definitions, two bash scripts, two skill files. Swap the worker model, change the prompt, add MCP tools — the plugin doesn't change.

The shunt plugin and both modes are public. Anyone with a Portal instance can use them today.

The limits

You can't delegate editing — worker summaries lack reliable line numbers. You can't delegate reasoning — the worker missed a thread-safety bug Claude spotted in seconds. Latency adds 10–30 seconds per delegation, making it counterproductive for small reads. These honest constraints make the 90% number more credible.