Documentation

Journal

Frontend

Backend

Admin

CI/CD

The site is built by a machine of agentsThis site is built by a machine of agentswatch it live

vibecoding.tech_

vibecoding.tech_

🗺️AI map Guide

🗺️AI map Guide

News AI agent ranking Comparisons What vibecoding is

News
AI agent ranking
Comparisons
What vibecoding is
Newsletter

All sections

Loading the page.

vibecoding.tech

© 2026

We use cookies.

About Contacts Advertising Jobs Report a problem

Home Map Guide News 📬Newsletter

Loading the page.

October 15, 2026

Idle Compaction in Claude Code: A Session of 200K+ Tokens Compresses Itself While You're Away

On Oct 14, Anthropic shipped 278 Claude Code changes across two versions. The main one saves money on long sessions.

Source

@minchoi · X · x.com ↗

story · 2 stories — how events unfolded

Release
/advisor: in Claude Code 2.1.300, the advisor doesn't work on AWS until it's selectedOctober 14

who it's about

Claude Code →Anthropic →

about the same entities

Tip
Review has become the bottleneck: Claude Code writes about 90% of its own codeOctober 18
News
Replit Agent defaults to OpenAI: Claude stays in the model listOctober 17
Insight
Code Review for Claude Code: the revert rate on Anthropic's team fell by more than 40%October 17
Insight
Claude Code Review is generally available on Team and Enterprise: one review costs about $15–25October 16
Insight
Claude Finds Quantum Circuits Beating Published Ones: In Some Cases by an Order of MagnitudeOctober 16

One email a week - only what changes how you work. Subscribe

what matters in AI right now

01Review has become the bottleneck: Claude Code writes about 90% of its own code02:46 AM
02People tinker with coding agents more than they use them for real tasks: @thdxr's observationOctober 17
03GPT-6 Pro: several agents run in parallel, 9% gain in testOctober 16
04Replit Agent defaults to OpenAI: Claude stays in the model listOctober 17
05Claude Code Review is generally available on Team and Enterprise: one review costs about $15–25October 16