An in-depth interpretation of the Anthropic Developer Conference: Claude Code transforms into engineering infrastructure

Core conclusion

On May 6, 2026, Anthropic held its second developer conference in San Francisco. Although the conference did not release a new model, it released three heavyweight signals:

• Reached computing power cooperation with SpaceX, taking over all 220,000 GPUs in the Colossus 1 data center, and 300 MW of new capacity will be online within one month
• Claude Code has been upgraded from an AI coding assistant to an engineering automation infrastructure, with the entire usage limit doubled and the current limit cancelled.
• Managed Agents launches the "Dreaming" capability, allowing AI agents to achieve cross-session self-learning

These changes mean actual cost reduction and capability improvement for teams that use AI tools for content automation, software development, and agent deployment.

Key Points

• Time of incident: 2026-05-06 -Affected objects: Claude Code developers, AI Agent engineers, content automation team
• Core changes: Alleviation of computing power bottlenecks, comprehensive upgrade of tool chains, and evolution of agents from executors to collaborators

Background: Anthropic computing power crisis and unexpected alliance with SpaceX

Ami Vora, chief product officer of Anthropic, got straight to the point: "We won't release new models today, we will only talk about products."

Behind this sentence is the grim reality: Anthropic's API calls have increased 17 times year-on-year in the past year, and the original computing power infrastructure has long been overwhelmed. Musk has previously publicly called Anthropic "misanthropic and evil (anti-human and evil)", and the relationship between the two parties has become tense.

However, on the day of the conference, Musk posted on X that he was "deeply impressed" after in-depth contact with Anthropic's senior management and believed that Claude had a "high probability" of being beneficial to mankind.

Core Terms of the SpaceX Cooperation Agreement

This cooperation is a double-edged sword for Anthropic - it solves the immediate production capacity crisis, but Musk essentially holds a "self-destruct switch."

Comprehensive upgrade of Claude Code: from coding assistant to engineering infrastructure

The core product launch of this conference is focused on Claude Code - the core engine that has driven Anthropic's growth in the past year.

Usage limit doubled across the board

With the cancellation of peak period throttling, the Claude Opus series API speed has been greatly increased, and the upper limit of token input per minute for some levels has been increased by more than 15 times.

Centralized iteration of more than ten functions

Claude Code's product direction has shifted from "AI programming assistant in the terminal" to "asynchronous automation infrastructure for engineering teams":

• Unattended PR pipeline: From code generation, asynchronous review, CI repair to automatic merging, the entire process can be automated
• advisor strategy: Let the stronger Opus model provide advice on demand for lightweight models such as Sonnet to achieve higher scores and lower costs
• Third Party Integration: Customer eve uses this to obtain flagship model quality output at 5 times lower cost

Mercado Libre (which has 23,000 engineers) said it aims to automate 90% of its coding by Q3 this year.

Managed Agents Dreaming: Cross-session learning of agents

If the update of Claude Code is a horizontal expansion of engineering capabilities, then the "Dreaming" capability of Managed Agents points to a deeper architectural evolution.

Dreaming allows the agent to shift from "starting from scratch every time" to "accumulating learning across sessions" - this is a key leap in the evolution of AI from a tool to a collaborator.

Domestic Manus, Byte Coze, and Alibaba intelligent platform are all chasing the "Agents as Infrastructure" narrative, but in the dimension of cross-session self-learning, Anthropic's Dreaming is the most systematic attempt to make a public debut.

Special layout in the financial services field

The conference also announced the in-depth layout of the financial services field:

• 10 new Cowork and Claude Code plug-ins
• Microsoft 365 suite integration (Word, Excel, PowerPoint, etc.)
• MCP applications for financial services and insurance institutions

This is a signal for Anthropic to penetrate deeply into regulatory compliance industries, and is of great significance to teams engaged in financial technology and insurance technology automation.

Related extended information

• 36：Anthropic

Tool entry

AI tool names that appear naturally in the text will match the maintained tools library on the platform side: Claude Code, Claude, OpenAI, ChatGPT, DeepSeek, Gemini, Manus, Coze

Internal link guidance

• Want to learn how to use Claude Code to automate content? See: Claude Code automated writing practice: build an AI content production pipeline in 30 minutes
• Someone used Claude Code to start a business in 48 hours and earn monthly income $9,000: Claude Code 48 hours to start a business: one person + US$29 monthly fee, monthly income in 3 months $9,000
• How to use DeepSeek to replace Claude Code and save 90% of the cost: DeepSeek V4 vs Claude Code: 90% Cheaper, Same Quality