New Claude Code funsies

Opus 4.6 came out this week (not the Sonnet 5 that was promised… a guy can still dream) and Claude Code got a bunch of new features - some publicised and some less so.

Agent Swarms

Mike Kelly actually discovered this feature a few days before it came out; it was already in the Claude Code build but hidden behind a secret feature flag.

I managed to unlock a crazy new hidden feature in Claude Code called Swarms. You're not talking to an AI coder anymore. You're talking to a team lead. The lead doesn't write code - it plans, delegates, and synthesizes. When you approve a plan, it enters a new "delegation mode"

Mike Kelly @NicerInPerson · 24 Jan 2026

It’s now available generally but is behind a feature flag that needs to be enabled first. The docs also warn that it’ll eat through your token limit and that subagents are enough for most people.

Memory

Claude Code can now write things to its memory and will do so unprompted:

Compact then implement

In Plan mode, once the plan is devised Claude will offer to ‘compact chat and implement plan’, which clears out all the conversation gubbins you had beforehand and gives the agent more tokens to work with (as your context window isn’t polluted with a bunch of gabbing).

New words

Now Claude will say it’s “scampering”, “galloping”, “churning” and of course…

Anyone interested in a Claude Code “wibbling…” hoodie?