Issue #1

February 10, 20265 min read

Claude Opus 4.6 now in Claude Code with agent teams, auto memory

Claude Opus 4.6 is now available in Claude Code, bringing experimental agent teams, automatic memory, and fast mode support, alongside expanded hook events and PDF page-range reading. Coverage this week spans Goldman Sachs deploying Claude-powered agents, Apple adding Claude to Xcode, and a reported $20 billion Anthropic funding round.

The Week Anthropic Tanked the Market and Pulled Ahead of Its Rivals — Wall Street Journal

On this page

Highlight
Worth Knowing
Under the Hood
From Anthropic
In the News

Highlight

Claude Opus 4.6 Arrives With Agent Teams and Automatic Memory

Opus 4.6 is now available in Claude Code, bringing two significant new capabilities alongside it. You can enable experimental agent teams -- where multiple Claude instances collaborate in parallel -- by setting CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1, though be aware this is token-intensive. Claude also now automatically records and recalls memories as it works, reducing the need to re-explain context across sessions.

Worth Knowing

Fast Mode Now Supported on Opus 4.6

You can now use fast mode with Opus 4.6, bringing lower-latency responses to the most capable model. A fix also landed to ensure /fast becomes available immediately after enabling /extra-usage, rather than requiring a restart.

Multi-Agent Workflows Gain New Hook Events and Memory Scoping

If you're building with agent teams, three additions expand what you can do: TeammateIdle and TaskCompleted hook events let you react to agent lifecycle changes, Task(agent_type) syntax in frontmatter lets you restrict which sub-agents a given agent can spawn, and the new memory frontmatter field gives agents persistent memory scoped to user, project, or local.

Session Picker Now Shows Git Branch and Supports Remote Resume

Your session picker now displays the git branch and message count for each session, and you can search by branch name to find the right context quickly. OAuth users can also browse and resume sessions directly from claude.ai.

PDF Handling Upgraded With Page Ranges and Lightweight References

You can now read specific page ranges from PDFs using the pages parameter in the Read tool (e.g., pages: "1-5"), which helps avoid hitting size limits on large documents. PDFs over 10 pages now return a lightweight reference when @ mentioned rather than being dumped into context wholesale.

MCP OAuth Credentials and New Debug Command Added

You can now pass --client-id and --client-secret to claude mcp add for MCP servers like Slack that don't support Dynamic Client Registration. A new /debug command also lets Claude help you diagnose issues with the current session directly.

Under the Hood

Sandbox Permission Bypass Fixed, Along With Session and API Bugs

A security fix closes a gap where commands excluded from sandboxing via sandbox.excludedCommands or dangerouslyDisableSandbox could bypass the Bash ask permission rule when autoAllowBashIfSandboxed was enabled. Several other fixes landed too: PDF size errors no longer permanently lock sessions, temperatureOverride is now respected in the streaming path, and the prompt cache now correctly invalidates when tool schemas change.

68% Memory Reduction for Users With Many Sessions

If you use --resume frequently or have accumulated many past sessions, memory usage is down 68% thanks to a shift to lightweight stat-based loading with progressive enrichment instead of loading the full session index upfront.

From Anthropic

Anthropic Engineering · 8 min read

Building a C compiler with a team of parallel Claudes

Anthropic's engineering team walks through building a C compiler by running multiple Claude instances in parallel -- a hands-on look at the agent teams approach that shipped in experimental form this week. Worth reading alongside the new CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS feature.

Anthropic Engineering · 8 min read

Building a C compiler with a team of parallel Claudes

Claude Blog · 5 min read

Behind the model launch: What customers discovered testing Claude Opus 4.6 early

Four customer teams share what they found during early access to Opus 4.6 -- useful context on where the model performs well and what shaped the release you're now running.

Claude Blog · 5 min read

Behind the model launch: What customers discovered testing Claude Opus 4.6 early

Four customer teams share what they found during early access to Opus 4.6 -- useful context on where the model performs well and what shaped the release you're now running.

In the News

Wall Street Journal · 6 min read

The Week Anthropic Tanked the Market and Pulled Ahead of Its Rivals

The WSJ traces how Anthropic's focus on coding and enterprise clients drove a market-moving week -- useful background on the commercial trajectory behind the tools you use daily.

Wall Street Journal · 6 min read

The Week Anthropic Tanked the Market and Pulled Ahead of Its Rivals

The WSJ traces how Anthropic's focus on coding and enterprise clients drove a market-moving week -- useful background on the commercial trajectory behind the tools you use daily.

Reuters · 3 min read

Goldman Sachs teams up with Anthropic to automate banking tasks with AI agents, CNBC reports

Goldman Sachs is deploying Claude-powered agents to automate trade accounting and client onboarding -- a concrete example of the agentic workflows that agent teams in Claude Code are designed to enable.

Reuters · 3 min read

Goldman Sachs teams up with Anthropic to automate banking tasks with AI agents, CNBC reports

CNBC · 4 min read

Apple adds agents from Anthropic and OpenAI to its coding tool

Apple is bringing agentic coding to Xcode with Claude as one of the supported backends -- expanding where Claude Code's capabilities show up in developer workflows.

CNBC · 4 min read

Apple adds agents from Anthropic and OpenAI to its coding tool

Apple is bringing agentic coding to Xcode with Claude as one of the supported backends -- expanding where Claude Code's capabilities show up in developer workflows.

Reuters · 3 min read

Anthropic's more than $20 billion funding to close as soon as next week, Bloomberg News reports

A funding round of more than $20 billion is expected to close shortly, alongside an employee tender offer at a $350 billion valuation -- signaling the resources behind Anthropic's current development pace.

Reuters · 3 min read

Anthropic's more than $20 billion funding to close as soon as next week, Bloomberg News reports

With Opus 4.6, agent teams, and automatic memory all landing in the same week, there's a lot to explore -- the engineering post on building a C compiler with parallel Claudes is a good place to start if you want to understand what multi-agent workflows look like in practice.