
What happened
A developer writing on Zenn says he stopped dividing work to keep Claude Code's context small, letting Sonnet handle investigation through testing in one session, then running /clear.
Why it matters
He argues that interrupting a task just to save context breaks continuity, while a separate Opus session for review gives a fresh perspective on the code.
Summaries like this, in your inbox every morning.
The developer's shift came after noticing that even a single small request could balloon context, because recent Sonnet versions investigate related code, implement, verify, and test in one go. Splitting tasks further to save context meant interrupting that flow and re-explaining the situation in a new session, which he judged worse for keeping the reasoning behind implementation decisions intact.
His current setup separates concerns by session rather than by context size: Sonnet carries a task from request through testing in the main session, Opus reviews in a separate session to avoid being anchored by the implementation conversation, and fixes go back to the original Sonnet session until review passes. Only then does he run /clear. He explicitly notes this is his own judgment, not a quantitative comparison of quality, and that manual /compact is not something he has found necessary in his style of work.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Anthropic's internal study shows junior new hires in technical departments fell from 27 percent to 9 percent s…

Mastercard is testing a new intelligence layer in its Agent Pay Trust Framework in the United States that prod…

In a decision dated September 24, 2026, the Aachen Regional Court banned Google from spreading three false AI…

Three designs for turning encoded inputs into predictions: pooling all inputs into one shared vector, applying…

Across 24 counted runs of /code-review on a 41-line shopping-cart file with six planted bugs, --max-findings 1…

The tutorial, based on official Claude Code documentation and checked as of October 18, 2026, says failed fixe…
