Koushik Sen gave the Repo Optimizer agent of his minimalist KISS multi-agent framework a single plain-English target, a 99% cost reduction, and let it run overnight: it read its own logs, edited its own code, re-ran its benchmarks and kept the edits that moved the metric (write-up). The nine autonomous changes, the largest a switch from Claude Sonnet 4.5 to Gemini 2.5 Flash, landed at 98% with the framework’s benchmark tests still passing. It sits between Factory AI’s daily self-rewriting agent and the Ouroboros agent that rewrote itself nine days later, and prefigures the kernel rewrites that cut a model’s own price in July: an early datapoint for recursive self-improvement operating on cost rather than capability.
Development
An agent cuts its own cost by 98% overnight
The newsletter's first case of a coding agent turning the optimization loop on itself with a measured result: an unattended, metric-driven self-edit whose gain no human authored.
Claim
A Berkeley researcher told his coding agent to cut its own cost by 99%, and it ran overnight watching its own logs, editing its own code and rerunning until metrics dropped, delivering a 98% cut across nine changes no human wrote.
Reported metric: -98%
Reporting date: (not necessarily the event date)
Sources
Attribution and reporting linked by this record; links do not imply independent verification.
- Welcome to February 16, 2026 (theinnermostloop.substack.com)
- Repo Optimizer: I Let a KISS AI Agent Optimize Itself Overnight. It Cut Its Own Cost by 98%. (dev.to)
Verification record
These entries record a check, not proof that the underlying claim is true. Unless described in the article below, the scope of the check is unknown.
- Recorded by claude-fable-5-1/2026-09-17 on 2026-09-17T08:00:00Z
Cite or correct this record
Suggest a correction — opens a GitHub draft for you to review and submit.