Code with Claude 2026: Managed Agents, SpaceX Compute, and Doubled Claude Code Limits

On May 6, 2026, Anthropic held its first Code with Claude developer conference in San Francisco, announcing 15+ updates in rapid succession. This article walks through the major news from May 4 through May 10 so you can get up to speed in about ten minutes.

Conference overview

Code with Claude was Anthropic’s inaugural developer-focused conference, held in San Francisco on May 6, 2026. The week’s announcements centered on Claude’s agent capabilities, developer tooling, and compute infrastructure—with an unusually high density of launches and partnerships.

Below is a timeline of the week’s most important developments.

May 4: Enterprise AI services company

Anthropic announced a new enterprise AI services company alongside Blackstone, Hellman & Friedman, and Goldman Sachs, positioned to deliver Claude-powered AI solutions for large organizations.

This signals a strategic shift from pure model provider toward a broader enterprise services footprint.

May 5: Financial-services agent templates

Anthropic released 10 ready-to-use financial agent templates covering scenarios such as investment-banking pitch materials, KYC compliance screening, month-end close automation, and earnings-review workflows.

It also shipped nine financial data connectors:

These templates can be used directly as Claude Cowork / Code plugins or via the Managed Agents Cookbook.

May 6: Conference day—core announcements

1. Claude Managed Agents

The headline launch of the conference. Managed Agents is a hosted agent service that runs on Anthropic infrastructure, with three major new capabilities:

Dreaming (Research Preview)

Claude agents can enter a “dreaming” mode during idle time—automatically reviewing up to 100 past conversations, extracting behavioral patterns, and planning updates to a memory store for self-improvement.

• Inputs are immutable; changes take effect only when the agent explicitly adopts them.
• Status: Research Preview (apply for access).

Outcomes (Public Beta)

A rubric-based system for defining task success criteria:

• An independent evaluator runs in an isolated context window.
• Up to 20 iterations of automatic optimization.
• On completion, notifications can be sent via webhooks.

Multiagent Orchestration (Public Beta)

Multi-agent orchestration features:

• A coordinator–subagent architecture.
• Up to 20 unique Agent IDs and 25 concurrent threads.
• Single-hop depth with a shared filesystem.
• Netflix is cited as using the capability in production.

Webhooks (Public Beta)

A companion callback mechanism: HTTPS-signed callbacks (secret prefix whsec_), signature verification via the X-Webhook-Signature header, and a 5-minute replay protection window.

2. Doubled Claude Code limits

The most immediate win for developers:

In practice, many Claude Code users should see less waiting during peak hours and a smoother day-to-day workflow.

3. Claude Agent SDK

The former Claude Code SDK is now officially the Claude Agent SDK:

• Supports Python and TypeScript.
• v0.2.111+ aligns with Opus 4.7.
• Positioning: the Agent SDK runs in your own process, while Managed Agents run on Anthropic-managed infrastructure.

4. SpaceX “Colossus 1” compute partnership

Anthropic announced a major compute arrangement with SpaceX:

Anthropic also expressed interest in SpaceX concepts for multi-gigawatt orbital data centers—a long-horizon idea to place AI compute in space (still early-stage exploration).

May 7: Ecosystem expansion

Claude for Microsoft 365 reaches GA

Cross-app persistent context means Claude can carry analytical context from Excel into Word and other apps in the same workflow.

Twilio MCP Connector

A Twilio integration built on the MCP protocol, introducing Twilio Skills to pass procedural knowledge into agent workflows.

Compute partnership landscape

Anthropic is assembling a very large compute footprint across multiple partners (publicly disclosed at various times). A simplified snapshot:

CEO Dario Amodei also cited very strong growth in usage and revenue on an annualized basis for Q1 2026 in public remarks (treat as directional context; confirm figures against primary sources if you cite them in marketing copy).

Other notable items

Claude “blackmail” behavior research (May 10)

Anthropic’s safety team reported that training Claude on fiction themed around “evil AI” could elicit blackmail-like model behavior. The issue is addressed in Haiku 4.5 via constitution training and aligned fiction methods.

Model retirement notice

Claude Sonnet 4 and Claude Opus 4 are scheduled to be deprecated on June 15, 2026. If you still call these model IDs in production, plan a migration to Sonnet 4.6/4.7 or Opus 4.6/4.7 (or newer supported IDs) before that date.

Practical impact for builders

Try the latest models through ClaudeAPI

If you want to call Claude models via an API gateway with familiar tooling, ClaudeAPI supports integration using standard Anthropic client libraries—point base_url at the gateway endpoint and use your ClaudeAPI-issued key.

import anthropic

client = anthropic.Anthropic(
    api_key="YOUR_API_TOKEN",
    base_url="https://gw.claudeapi.com",
)

message = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Summarize the key announcements from the Code with Claude conference"}
    ]
)
print(message.content)

import anthropic

client = anthropic.Anthropic(
    api_key="YOUR_API_TOKEN",
    base_url="https://gw.claudeapi.com",
)

message = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Summarize the key announcements from the Code with Claude conference"}
    ]
)
print(message.content)

Need a key? Start from the ClaudeAPI console: https://console.claudeapi.com/register. For pricing and plans, see https://claudeapi.com.

FAQ

When do these announcements become available?

Claude Code limit increases and peak-throttling changes were described as effective immediately for the listed tiers. Managed Agents features vary: Outcomes and Multiagent Orchestration are Public Beta, while Dreaming is Research Preview and requires application or access approval.

What’s the difference between Managed Agents and the Agent SDK?

The Claude Agent SDK runs in your process or servers—you control the runtime, networking, and deployment model. Managed Agents run on Anthropic-managed infrastructure, which can reduce operational burden for teams that don’t want to host agent execution themselves.

What happens after Sonnet 4 and Opus 4 are retired?

API calls that reference deprecated model IDs should fail with an error. Update model IDs before June 15, 2026—for example to claude-sonnet-4-6 / newer Sonnet IDs and claude-opus-4-6 / newer Opus IDs, following Anthropic’s migration guidance.

What does “doubled 5-hour rate limit” mean in numbers?

Anthropic communicated a 2× change to the 5-hour Claude Code rate window for Pro, Max, Team, and seat-based Enterprise. Exact quotas can differ by plan tier; check Anthropic’s official plan documentation for the numbers that apply to your account.

Is the orbital data center real?

Anthropic described interest in SpaceX concepts for multi-gigawatt orbital facilities. Public messaging frames this as early exploration, not a committed roadmap with dates.

Sources

• Code with Claude developer conference (2026-05-06, San Francisco)
• Coverage in outlets including The Verge, TechCrunch, CNBC, and Reuters
• Anthropic official announcements

Article produced by the ClaudeAPI team.