Doubled Code limits, but with a twist — AWS GA, Agent View, /goal

TLDR;

A massive week for builders: Claude Platform on AWS went GA on May 11, agent view landed in Claude Code, and three breaking changes hit the TypeScript Agent SDK (v0.3.142) that require migration. Claude Code weekly limits jumped another 50% (stacking with last week's 2× 5-hour bump), Opus 4.7 fast mode entered research preview, and a new programmatic credit pool ($20–$200/month depending on plan) lands June 15 — claim window opens June 8. Five daily Claude Code releases (v2.1.139–v2.1.143) shipped claude agents, /goal, projected context cost in the marketplace, and plugin dependency enforcement. Claude for Legal opened 20+ MCP connectors and 12 practice-area plugins as Apache-2.0. Plus Cursor 3.4 ships cloud agent environments, GitHub launches a Copilot desktop app, and the Bun-in-Rust rewrite is at 99.8% test pass.

Your AI shouldn't grade its own homework

Claude Code writes beautiful code. So does Codex. But here's the thing — they also think they write beautiful code. And when you ask an AI to review code it just wrote, you get the intellectual equivalent of a student grading their own exam. Shockingly, they always pass.

CodeRabbit CLI plugs into Claude Code and Codex as an external reviewer — different AI agent, different architecture, 40+ static analyzers, and zero emotional attachment to the code it's looking at. The agent writes, CodeRabbit reviews, and the agent fixes. Loop until clean.

You show up when there's actually something worth approving.

One command. Autonomous generate-review-iterate cycles. The AI still does the work. It just doesn't get to decide if the work is good anymore.

Free tier available. Try CodeRabbit's CLI →

API & Model Updates

Claude Code weekly limits up 50% through July 13

Effective immediately, weekly usage limits for Claude Code are 50% higher across CLI, IDE extensions, desktop, and web — no opt-in required — for all Pro, Max, Team, and seat-based Enterprise users. This stacks with the 2× increase to 5-hour limits announced the week prior.

Fast mode for Claude Opus 4.7 in research preview — API waitlist open

Opus 4.7 fast mode (~2.5× higher output speed at full Opus 4.7 intelligence) is now available in research preview on the API and in Claude Code; it's also live in Cursor, Emergent Labs, Factory AI, v0, Warp, and Windsurf. API users need to join the waitlist; fast mode now defaults to Opus 4.7 (previously 4.6) — set CLAUDE_CODE_OPUS_4_6_FAST_MODE_OVERRIDE=1 to pin to 4.6.

New programmatic credit pool — claim by June 8, live June 15

Starting June 15, paid Claude plans get a dedicated monthly credit ($20–$200 depending on plan) for programmatic usage — covering the Claude Agent SDK, claude -p, Claude Code GitHub Actions, and third-party apps built on the Agent SDK. Interactive usage limits are unchanged. You'll receive an email on June 8 to claim; no action needed before then. No API key needed for Agent SDK usage after June 15. Third-party tools like Conductor and OpenClaw draw from this credit pool.

Claude Platform on AWS — generally available (May 11)

AWS customers now get the full Claude Platform feature set — Claude Managed Agents, prompt caching, code execution, web search, skills, and all new betas on day one — with AWS IAM auth, CloudTrail audit logging, and a single AWS invoice. Pricing and rate limits are identical to the direct Claude API. Claude on Amazon Bedrock remains available (AWS as data processor) but does not yet include the agent-platform surface.

Introducing Claude for Small Business (May 13)

Anthropic launched a Claude plan tier aimed at small businesses. Builders should note this as a new addressable customer segment for products built on Claude.

Claude Code & CLI

Claude Code v2.1.139: agent view, /goal command (May 11)

Two major feature additions: agent view (claude agents — Research Preview) gives you a single list of every Claude Code session across running, blocked, and done states with inline reply to unblock; /goal lets you set a completion condition and Claude keeps working across turns until it's met, showing live elapsed/turns/tokens. Also adds args: string[] for exec-form hooks (no shell quoting issues) and continueOnBlock for PostToolUse hooks.

Claude Code v2.1.140: Agent tool subagent_type matching, settings hot-reload fixes (May 12)

Agent tool subagent_type matching now accepts case- and separator-insensitive values ("Code Reviewer" resolves to code-reviewer). Fixes a regression in settings hot-reload with symlinked settings files, claude --bg mid-request connection drops, and a recurring event-loop stall on Windows from missing executables.

Claude Code v2.1.141: terminalSequence in hooks, /feedback with session history, rewind summarize (May 13)

Hooks can now emit desktop notifications, window titles, and bells via terminalSequence without a controlling terminal. /feedback can attach recent sessions (last 24h or 7 days). The rewind menu gains "Summarize up to here" to compress earlier context. CLAUDE_CODE_PLUGIN_PREFER_HTTPS clones GitHub plugin sources over HTTPS for environments without a GitHub SSH key.

Claude Code v2.1.142: claude agents flags, fast mode defaults to Opus 4.7 (May 14)

claude agents gains --add-dir, --settings, --mcp-config, --plugin-dir, --permission-mode, --model, --effort, and --dangerously-skip-permissions flags to configure dispatched background sessions. Fixes MCP_TOOL_TIMEOUT not raising the per-request fetch timeout (which capped tool calls at 60s regardless of config), and background sessions disappearing after macOS sleep/wake.

Claude Code v2.1.143: plugin dependency enforcement, projected context cost, worktree bgIsolation (May 15)

claude plugin disable now refuses when another enabled plugin depends on the target (with a copy-pasteable disable-chain hint); claude plugin enable force-enables transitive dependencies. The /plugin marketplace browse pane shows projected per-session context cost. New worktree.bgIsolation: "none" setting lets background sessions edit the working copy directly for repos where worktrees are impractical. Fixes stop hooks that block repeatedly looping forever (capped at 8 consecutive blocks; override via CLAUDE_CODE_STOP_HOOK_BLOCK_CAP).

/goal, stop hooks, and auto mode — how to keep Claude working until done

The @ClaudeDevs account posted a useful thread on long-running agent patterns: /goal sets a completion condition; a stop hook gives programmatic control over when Claude is allowed to finish (run your test suite, hit a CI endpoint); auto mode (shift+tab in CLI or mode selector on desktop) means Claude doesn't wait on you between turns.

Prompt cache pre-warming tip: cut time-to-first-token on long prompts

Send your system prompt before the user prompt — Claude writes it to cache but skips generating output. When the real user request lands, it hits a warm cache. See the prompt caching docs for implementation details.

Agent SDK & Managed Agents

Claude Agent SDK TypeScript v0.3.142: three breaking changes (May 14)

Action required before upgrading. Three breaking changes: (1) The v2 session API (unstable_v2_createSession, unstable_v2_resumeSession, unstable_v2_prompt, SDKSession, SDKSessionOptions) is removed — migrate to query(), pass an AsyncIterable for multi-turn or options.resume to continue a session. (2) MCP servers now connect in the background by default; set MCP_CONNECTION_NONBLOCKING=0 to restore old behavior, or alwaysLoad: true to require a server in turn 1. (3) Headless and SDK sessions now use Task tools (TaskCreate / TaskUpdate / TaskGet / TaskList) instead of TodoWrite — accumulate by task ID instead of replacing a snapshot list.

Claude Agent SDK TypeScript v0.3.143: peer dependencies (May 15)

@anthropic-ai/sdk and @modelcontextprotocol/sdk are now peerDependencies instead of dependencies. Runtime is unaffected (both are bundled); npm/bun/pnpm users get auto-install. Yarn classic users should add them explicitly for full TypeScript type resolution.

Claude Agent SDK Python v0.2.82: MCP background connect, Task tools replace TodoWrite (May 15)

Mirrors the TypeScript v0.3.142 breaking changes: MCP servers connect in the background by default, and Task tools replace TodoWrite. Also adds a public EffortLevel type alias ("low", "medium", "high", "max", "xhigh") exported from the package root; fixes a stderr callback isolation bug where a raising callback would silently drop all subsequent stderr lines.

Build and deploy agents at scale with Claude Managed Agents on AWS

Claude Managed Agents — the hosted agent-loop service with features like the advisor strategy, code execution, and web search — is now available through the Claude Platform on AWS. Anthropic operates the service; all new features ship the same day they go live on the native Claude API.

MCP Ecosystem

Claude for Legal: 20+ MCP connectors, 12 plugins, open-sourced (May 12)

Anthropic released 20+ MCP connectors for legal software (Ironclad, DocuSign, Relativity, Everlaw, iManage, NetDocuments, Harvey, Thomson Reuters CoCounsel, and more) plus 12 practice-area plugins (Commercial, Corporate, Employment, Privacy, Product, Regulatory, IP, Litigation, and others). Four plugins are available as "cookbooks" deployable as Managed Agents through the Claude API. Everything is Apache 2.0 licensed at github.com/anthropics/claude-for-legal.

Developer Tools & Community

Claude's Constitution now an audiobook — read by authors Amanda Askell and Joe Carlsmith (May 11)

The audiobook includes a Q&A on the writing process, the philosophies that shaped the document, and how it might change as models become more capable.

Bun rewritten in Rust by Anthropic/Bun team — 99.8% test suite pass, blog post coming

Jarred Sumner (Anthropic, Bun) confirmed a 960,000-LOC Rust rewrite of Bun passes 99.8% of the existing test suite across Linux, Windows, and macOS; no regressions found on benchmarks vs. the Zig implementation. The team is dogfooding on Claude Code internally before external release; blog post expected soon. This is relevant to Claude Code builders since Bun is Claude Code's runtime.

Competitor Dev Tools

Cursor 3.4: Cloud agent development environments, 70% faster cached builds (May 13)

Cursor ships multi-repo cloud agent environments with Dockerfile-based config, build secrets scoped to build steps only, 70% faster layer-cached rebuilds, per-environment version history with rollback, audit logs, and environment-scoped egress/secrets controls. Relevant for teams evaluating parallel agent infrastructure that works alongside Claude Code.

GitHub Copilot App: technical preview with parallel sessions and Agent Merge (May 14)

GitHub's new desktop app supports parallel Copilot agent sessions (each with its own worktree and branch), a GitHub inbox across repos, and Agent Merge — which watches CI, downloads failure logs, resolves conflicts, and pushes fixes automatically. New individual plan sign-ups are paused; Opus models removed from Copilot Pro (remain in Pro+).

OpenAI Codex comes to iOS and Android — all plans (May 14)

Codex is now available in preview in the ChatGPT mobile app, letting users start or continue threads, approve actions, and review diffs and test results from their phone while Codex runs on a connected Mac host.

Enterprise & Business

PwC expanded partnership: deploying Claude to build technology and execute deals (May 14)

PwC is deploying Claude to build technology, execute transactions, and reinvent enterprise functions for clients. Builders working in professional services or building on top of Claude for enterprise workflows should note PwC as both a reference customer and potential channel.

Anthropic + Gates Foundation: $200M partnership for global health, education, and agriculture (May 14)

Anthropic is committing $200M in grants, Claude credits, and technical support to Gates Foundation programs in global health, life sciences, education, agriculture, and economic mobility. Builders in these verticals may find partnership or grant opportunities through this channel.

2028: Two scenarios for global AI leadership (May 14)

Anthropic published a policy paper on US vs. China AI competition, arguing the US and democratic allies hold the frontier lead today and outlining what it will take to maintain it. Relevant context for builders whose products touch export controls, government contracts, or international deployment.

Action Items

Immediate:

API users wanting Opus 4.7 fast mode: join the waitlist now — it's research preview with separate limits

Before upgrading to Agent SDK TypeScript v0.3.142:

Migrate off unstable_v2_createSession / unstable_v2_resumeSession / unstable_v2_prompt to query()
Update any MCP server startup logic that assumes synchronous connection (set MCP_CONNECTION_NONBLOCKING=0 to restore old behavior)
Update tool consumers from TodoWrite snapshot replacement to Task tool accumulation by task ID (TaskCreate / TaskUpdate / TaskGet / TaskList)

By June 8, 2026:

Watch for the Anthropic email to claim your new programmatic credit pool — you must claim it before June 15

By June 15, 2026:

Programmatic credit pool goes live: Agent SDK, claude -p, Claude Code GitHub Actions, and third-party Agent SDK apps will draw from the new credit pool instead of interactive limits — verify your usage patterns won't exceed the new credit tier

Mert Duzgun