Issue #24

July 20, 20265 min read

Meta may lease compute to Anthropic for $10B

/fork now spawns background sessions in claude agents, session-wide caps curb runaway WebSearch, subagent, and long MCP calls, and a new EndConversation tool ends abusive sessions. Meanwhile, Meta is reportedly negotiating a $10B compute deal with Anthropic, Bloomberg covers AI's impact on software engineering, and DealBook reports an Anthropic IPO is in the works.

Meta in Talks to Lease Computing Power to Anthropic in Potential $10 Billion Deal — The New York Times

On this page

Highlight
Worth Knowing
Under the Hood
From Anthropic
In the News

Highlight

Fork Becomes a Background Session and Runaway Loops Get Hard Caps

/fork now copies your conversation into a new background session that shows up as its own row in claude agents, while the old in-session subagent behavior moves to /subtask. Alongside that, session-wide caps stop runaway agents: 200 WebSearch calls (CLAUDE_CODE_MAX_WEB_SEARCHES_PER_SESSION), 200 subagent spawns (CLAUDE_CODE_MAX_SUBAGENTS_PER_SESSION, reset by /clear), and MCP tool calls running longer than two minutes now move to the background automatically (CLAUDE_CODE_MCP_AUTO_BACKGROUND_MS). Together these change how you should think about parallelizing work and containing badly behaved agents.

Worth Knowing

Claude Can Now End Abusive Sessions

Claude Code gains the EndConversation tool that lets Claude terminate a session with highly abusive users or jailbreak attempts, matching behavior that has been on claude.ai since 2025. See Anthropic's research post for the rationale.

/resume Gains a Picker in the Agent View

Typing /resume in the agent view opens a picker of past sessions — including sessions you deleted from the list — and resumes your pick as a background session. Useful when you want to revive an old thread without leaving the agent dashboard.

Subagents Now Inherit the Parent's Permission Mode

The Task tool's mode parameter is deprecated and ignored; subagents inherit the parent session's permission mode by default. If you were relying on per-task mode overrides, expect subagents to behave like their parent going forward.

/verify and /code-review Are Now Manual Invocations

Claude no longer runs the /verify and /code-review skills on its own. Invoke them explicitly with /verify or /code-review when you want that behavior.

Screen Reader Mode and Vim Insert Remaps Ship

A new opt-in screen reader mode renders plain text for assistive tech — enable it with claude --ax-screen-reader, CLAUDE_AX_SCREEN_READER=1, or "axScreenReader": true in settings. Vim users also get a vimInsertModeRemaps setting to map two-key insert-mode sequences like jj to Escape.

Under the Hood

Bash and PowerShell Permission Checks Hardened

Version 2.1.214 closes several permission-check gaps: a bypass in Windows PowerShell 5.1 sessions, file-descriptor redirect forms bash parses differently, commands over 10,000 characters (which now always prompt), zsh subscripts in [[ ]], and certain help/man invocations that could run unsafe options. Plan mode also no longer auto-runs file-modifying Bash commands like touch and rm without a permission prompt.

Big Memory and Session-Size Wins in Long Sessions

Version 2.1.208 fixes several long-session memory leaks (MCP stdio stderr up to 64 MB per server, LSP documents now LRU-capped at 50, unbounded headless/SDK growth from large tool results), cuts session transcript size up to 79x in edit-heavy sessions by pruning superseded file-history backups, and bounds the edit read cache to 16 MB. Print/SDK sessions with many MCP tools also see up to 7x faster tool rounds from tool-pool caching.

From Anthropic

Anthropic Engineering · 4 min read

How Anthropic runs large-scale code migrations with Claude Code

A step-by-step guide to running large code migrations with Claude Code agents, including Bun's million-line Zig-to-Rust port. Directly applicable if you're planning any nontrivial refactor or language migration.

Anthropic Engineering · 4 min read

How Anthropic runs large-scale code migrations with Claude Code

· 14 min read

Zero risk isn't the job: a CISO's guide to agentic AI

Anthropic's Deputy CISO lays out a four-question framework for assessing agentic AI risk and the controls that keep agent deployments bounded and auditable. Useful reading if you're the one answering security questions about Claude Code inside your org.

· 14 min read

Zero risk isn't the job: a CISO's guide to agentic AI

In the News

The New York Times · 4 min read

Meta in Talks to Lease Computing Power to Anthropic in Potential $10 Billion Deal

Meta is reportedly in talks to lease compute to Anthropic in a deal potentially worth $10 billion. Another data point on the compute buildout underpinning the Claude models you use in Claude Code.

The New York Times · 4 min read

Meta in Talks to Lease Computing Power to Anthropic in Potential $10 Billion Deal

Meta is reportedly in talks to lease compute to Anthropic in a deal potentially worth $10 billion. Another data point on the compute buildout underpinning the Claude models you use in Claude Code.

Bloomberg · 8 min read

For Software Engineers, the AI Reckoning Is Already Here

Bloomberg's feature on how Anthropic and OpenAI tools are reshaping the profession of coding. Worth reading for the wider context around the tool you spend all day in.

Bloomberg · 8 min read

For Software Engineers, the AI Reckoning Is Already Here

Bloomberg's feature on how Anthropic and OpenAI tools are reshaping the profession of coding. Worth reading for the wider context around the tool you spend all day in.

Wired · 6 min read

Here's Why Anthropic Is Pushing States to Regulate AI Faster

Wired on Anthropic's push for faster state-level AI regulation and what's motivating the stance. Relevant if you're tracking how policy could shape Claude Code deployment options.

Wired · 6 min read

Here's Why Anthropic Is Pushing States to Regulate AI Faster

Wired on Anthropic's push for faster state-level AI regulation and what's motivating the stance. Relevant if you're tracking how policy could shape Claude Code deployment options.

The New York Times · 4 min read

Anthropic Inches Toward a Mega-I.P.O.

DealBook reports Anthropic is lining up investor meetings as an IPO nears. Useful context for anyone betting on Claude Code as a long-term platform.

The New York Times · 4 min read

Anthropic Inches Toward a Mega-I.P.O.

DealBook reports Anthropic is lining up investor meetings as an IPO nears. Useful context for anyone betting on Claude Code as a long-term platform.

A dense week: the /fork change and new subagent inheritance behavior are the ones most likely to affect your existing workflows, so check any scripts or agent configs that relied on the old semantics.