Note: This edition is assembled from available enterprise research and a combined weekly summary.

This was a week of platform milestones and infrastructure growth for Anthropic. The Claude Platform on AWS went GA โ€” bringing the full API feature set to AWS customers with IAM authentication and consolidated billing. Claude Managed Agents entered public beta with dreaming (self-improving memory), outcomes, and multiagent orchestration. The Agent SDK credit model (effective June 15) restores third-party harness access for subscription users, while Anthropic reported an 80-fold Q1 revenue increase to a $44 billion annualized run rate. Here is the Claude Weekly for May 14, 2026.

Claude Platform on AWS Goes GA

The Claude Platform on AWS launched in general availability on May 11, giving AWS customers the full Claude API feature set on Anthropic-managed infrastructure, accessible directly through AWS with IAM authentication and AWS billing (Source). This is a significant shift from the previous model, where AWS Bedrock was the sole path for AWS customers. Now teams get:

  • Messages API, Files API, and Message Batches API
  • Claude Managed Agents (beta), code execution, web search and fetch
  • Skills (beta), MCP connector (beta), prompt caching, and citations
  • Claude Console development environment
  • Models: Opus 4.7, Sonnet 4.6, and Haiku 4.5

The ant CLI also launched โ€” a command-line client for the Claude API with native Claude Code integration and YAML-based resource versioning (Source).

Why it matters for enterprise teams: AWS customers can now run Claude through their existing AWS procurement, compliance, and IAM workflows without a separate Anthropic account. For organizations with strict AWS-first policies, this removes the last procedural barrier to Claude adoption. Combined with the Claude Console, teams get a managed development environment alongside the AWS billing they already use.

Claude Managed Agents: Public Beta with Dreaming

Claude Managed Agents entered public beta with three major capabilities shipping simultaneously (Source):

  • Dreaming (research preview) โ€” agents review past sessions to identify patterns, mistakes, and preferences, then update their memory for self-improvement without explicit retraining. This addresses the context window ceiling for long-running multi-agent projects.
  • Outcomes โ€” declarative success criteria for agent sessions, letting teams define what “done” looks like and have agents work autonomously toward that target.
  • Multiagent orchestration โ€” coordinate multiple Claude agents on complex, multi-step tasks with webhooks for session and vault lifecycle events.

Memory is now in public beta with vault credential background refresh for mcp_oauth credentials. Per-tool permissions, managed credential vaults, and full audit logging are included.

Why it matters: Dreaming is the first practical implementation of persistent agent memory at platform scale โ€” agents that learn across sessions without manual retraining. For enterprise teams building autonomous workflows, this bridges the gap between stateless API calls and agents that get better the longer they run.

Agent SDK Credit Model: Third-Party Harnesses Return

Anthropic announced a new Agent SDK credit model effective June 15, 2026, that restores programmatic access for tools like OpenClaw, Pi, and Conductor that were cut off from subscription quota in April (Source). Every paid Claude plan now gets a separate monthly SDK credit for programmatic usage:

  • Pro: $20/month
  • Max 5x: $100/month
  • Max 20x: $200/month

Usage drains from credits first, then falls back to pay-as-you-go or pauses. Subscribers get a claim link email on June 8.

Why it matters: The April cutoff created tension in the developer ecosystem โ€” users who paid for Claude subscriptions found their preferred third-party tools blocked from using those same credits. This restores the subscription value proposition while giving Anthropic a metering lane for high-volume programmatic API use. For teams running OpenClaw, agent orchestrators, or CI/CD integrations with Claude, this is the green light to stay on subscription rather than switching to pay-as-you-go API keys.

$44B Annualized Run Rate, $900B Valuation Talks

CEO Dario Amodei reported 80-fold Q1 revenue growth to an annualized run rate of approximately $44 billion (Source). Over 1,000 enterprise customers now spend more than $1M/year with Anthropic โ€” doubled in under two months. Eight of the ten largest US companies use Claude.

Separately, reports indicate Anthropic is in talks to raise $40โ€“50 billion at an $850โ€“900 billion valuation, with investor demand so intense that some institutions offering $5B commitments could not secure allocation (Source). IPO speculation points to October 2026 as a potential timeline (Source).

The SpaceX/xAI Colossus 1 compute partnership provides 300+ MW of additional capacity, powering doubled Claude Code rate limits and removed weekly usage caps (Source).

Claude Security Public Beta

Claude Security entered public beta for Claude Enterprise customers, bringing AI-powered code vulnerability scanning and fix proposals powered by Opus 4.7 (Source). Features include scheduled and targeted scans, triage tracking, and integration with audit systems. Technology partners include CrowdStrike, Microsoft Security, Palo Alto Networks, SentinelOne, TrendAI, and Wiz.

Why it matters: Unlike most code scanning tools that stop at detection, Claude Security proposes fixes โ€” reducing the triage-to-remediation cycle. No API or custom agent build is required; if your organization already uses Claude Enterprise, scanning is available today. This competes directly with GitHub Advanced Security and GitLab SAST, but with the advantage of Claude’s reasoning to understand vulnerability context rather than pattern-match. For a broader look at AI-powered code risk, see our post on AI risk across codebases.

Claude Code: Agent View and Week 19 Release

Agent View (Research Preview) launched โ€” a single CLI view (claude agents) that shows every Claude Code session: running, blocked on input, or complete (Source). Users can start agents, send them to the background, peek at status, and resume when input is needed. Available from CLI, desktop app, VS Code, JetBrains, web, and Slack.

The Week 19 release (v2.1.128โ€“v2.1.136) also includes:

  • Plugins can load from .zip archives and URLs (--plugin-url)
  • Hard deny rules in auto mode โ€” block actions unconditionally
  • claude ultrareview subcommand for non-interactive CI/script usage
  • Sub-agent progress summaries now hit prompt cache, cutting cache_creation tokens by roughly 3x

Claude for Small Business launched May 13 โ€” a package of connectors and ready-to-run workflows for QuickBooks, Google Drive, Slack, and more (Source). A free 10-city Claude SMB Tour started May 14 in Chicago.

Financial services agents โ€” 10 pre-built templates deployed at JPMorganChase, Goldman Sachs, Citi, AIG, and Visa, with SOC compliance configurations and PII masking. FIS announced a partnership for financial crimes use cases (Source).

Legal ecosystem โ€” Over 20 MCP connectors and 12 practice-area plugins launched, connecting Claude with Thomson Reuters CoCounsel Legal, Harvey, and Solve Intelligence for litigation, contracts, discovery, and IP analysis (Source).

Enterprise pricing shift โ€” Claude Enterprise moved to consumption-based billing, removing bundled tokens from seat deals (Source). Heavy users face higher costs; Anthropic gains better demand management.

MCP and the Ecosystem

MCP 1.4 RC โ€” streaming context updates (server pushes changes to client proactively) is the headline feature (Source). Cross-provider routing is now common across Claude, GPT-5.5, DeepSeek V4-Pro, and 300+ models. Official registry servers reached ~520.

Adoption milestones: 78% of enterprise AI teams report at least one MCP-backed agent in production. 9,400+ public servers (7.8x YoY growth). 67% of CTOs name MCP their default agent-integration standard (Source).

Governance: Clare Liguori (Core Maintainer) and Den Delimarsky (Lead Maintainer) added to the MCP project, now under the Agentic AI Foundation (AAIF) (Source). As the ecosystem matures, AI governance frameworks become critical for enterprise teams deploying MCP-backed agents at scale.

Research: Teaching Claude Why and Petri Donation

Anthropic published significant safety and interpretability research this week:

  • Teaching Claude Why โ€” reducing agentic misalignment by teaching Claude the reasoning behind its actions (Source)
  • Natural Language Autoencoders โ€” translating internal numerical model representations into human-readable text for unprecedented visibility into reasoning (Source)
  • Petri open-source donation โ€” Anthropic donated its automated behavioral evaluation tool to the AI safety community (Source)
  • The Anthropic Institute focus areas published for formal safety and governance research (Source)

Also This Week

Claude for Excel, PowerPoint, and Word went GA with admin controls via AppSource. Outlook integration is in beta (Source).

Claude Opus 4.7 Fast Mode entered research preview โ€” set speed: "fast" for significantly faster output token generation at premium pricing (Source).

Creative tool integrations expanded โ€” connectors for Adobe, Blender, SketchUp, and Autodesk position Claude beyond coding into design and 3D workflows (Source).

Pentagon supply chain risk reversal โ€” the White House is developing guidance to work around the Pentagon’s supply chain risk designation, potentially reopening federal access (Source).

What to Watch

  • Agent SDK credit rollout (June 8โ€“15) โ€” All paid subscribers get a claim email on June 8; credits go live June 15. This reshapes the third-party tool economics for OpenClaw, Pi, and Conductor users.
  • Claude Opus 4.7 Fast Mode pricing โ€” Research preview waitlist is open. If fast mode proves viable at scale, it changes the economics of real-time agent applications.
  • MCP 1.4 GA โ€” Streaming context updates in RC. The first formal spec version since November 2025 could land in the coming weeks.
  • IPO timeline โ€” October 2026 speculation. Watch for formal S-1 signals this summer.
  • EU regulatory attention on Mythos โ€” Bank regulator warnings could affect European access and compliance requirements.

That is this week’s Claude Weekly. The AWS Platform GA removes a procurement friction point for enterprise teams, Managed Agents with dreaming deliver on the persistent-memory promise, and the Agent SDK credit model restores the subscription value proposition for developers running third-party tools. Platform maturity, infrastructure scale, and business fundamentals are all accelerating.

Put these capabilities to work in your organization. Whether you’re evaluating Claude on AWS, implementing Managed Agents for autonomous workflows, or planning an enterprise AI governance strategy, Big Hat Group can help. Contact our team or explore our AI & Automation services.

Check back next week for the latest.