Code with Claude 2026: Every Announcement from Anthropic's Biggest Event Yet

Anthropic’s Code with Claude 2026, held in San Francisco on May 6 with follow-up events in London (May 19) and Tokyo (June 10), was the company’s most significant product event to date. Five major features shipped, usage limits doubled across every paid plan, and a compute partnership with SpaceX gave Anthropic access to over 220,000 Nvidia GPUs.

Then on May 13, Anthropic launched Claude for Small Business, extending AI agents into QuickBooks, PayPal, HubSpot, and a dozen other tools that small businesses use daily.

This roundup covers every announcement, what each one means, and how it connects to the broader AI SEO shift.

The SpaceX compute deal: 220,000 GPUs

The foundation beneath every other announcement: Anthropic struck a deal with SpaceX to use the Colossus 1 data center in Memphis, over 220,000 Nvidia GPUs and 300 megawatts of power. This is one of the largest compute partnerships in AI history.

The immediate impact: doubled usage limits across every paid plan. Pro, Max, Team, and Enterprise users all received 2x their previous token limits. Peak-hour throttling was removed for Pro and Max accounts entirely.

For API users, the increases were even larger:

Tier 1: 1,500 percent increase in maximum input tokens per minute, 900 percent increase in output tokens per minute
Higher tiers: proportional increases across the board

The deal also includes an exploration partnership to develop orbital AI compute capacity, literally putting data centers in space to bypass terrestrial power constraints. That part is still exploratory, but the Memphis data center is operational now.

Why this matters: Anthropic’s biggest competitive weakness has been capacity constraints. Users hitting rate limits during peak hours, Claude Code sessions timing out during long tasks, and API customers facing throttling. The SpaceX deal addresses all of these at once. More capacity means more consistent performance, which means Claude becomes more viable for production workloads that cannot tolerate interruptions.

Doubled usage limits: what you actually get now

Here is the practical impact by plan:

Plan	Before	After	Peak Hours
Free	Limited	Unchanged	Still throttled
Pro ($20/mo)	Standard limits	2x limits	No peak throttling
Max ($100-200/mo)	5-20x Pro	2x previous	No peak throttling
Team ($25/user/mo)	Team limits	2x limits	Reduced throttling
Enterprise	Custom	2x limits	Custom SLAs

The removal of peak-hour throttling for Pro and Max is arguably more impactful than the raw limit increase. Previously, Claude would reduce output quality and speed during high-traffic periods (typically US business hours). That restriction is gone.

For developers using Claude Code for multi-hour autonomous coding sessions, this means fewer interruptions and more consistent performance during the workday when it matters most.

Dreaming: agents that learn from their own mistakes

The most technically ambitious announcement: Dreaming is a new capability where Claude agents review their own past sessions, extract patterns, and write notes that future agents can reference.

How Dreaming works

When a Claude Code agent completes a task, say, refactoring a module, it may have encountered errors, discovered undocumented patterns in the codebase, or found approaches that worked better than others. Without Dreaming, that knowledge disappears when the session ends.

With Dreaming, a scheduled background process reviews completed sessions and curates a memory store. The next agent that works on the same codebase can access these notes, getting up to speed faster and avoiding mistakes that previous agents already made.

Dreaming surfaces patterns that no single agent can see on its own:

Recurring mistakes across sessions
Workflows that agents converge on independently
Team preferences shared across multiple developers’ sessions
Codebase-specific knowledge that is not documented anywhere

Real-world impact

Harvey, the legal AI company, reported a 6x jump in task completion rates after enabling Dreaming for their Claude agents. The agents learned which approaches worked for their specific legal document formats and stopped repeating errors across sessions.

Dreaming is available in research preview. It represents a shift from stateless AI tools (every session starts from zero) to persistent AI agents that accumulate expertise over time.

Outcomes: defining what success looks like

Outcomes lets you write a rubric describing what a successful result looks like. A separate grader, running in its own context window, independent from the working agent, evaluates the output against your criteria.

When something does not meet the rubric, the grader pinpoints what needs to change and the agent takes another pass. This creates a feedback loop: the agent works, the grader evaluates, the agent improves, until the outcome criteria are met.

The key design decision: the grader runs in a separate context window so it is not influenced by the agent’s reasoning. It evaluates the output objectively, not the process that produced it.

For teams building AI-powered content workflows, Outcomes means you can define quality standards, “the meta description must be under 155 characters, include the primary keyword, and match the page’s actual content”, and the agent will iterate until the standard is met. This connects directly to how AI systems evaluate content quality.

Multi-agent orchestration: scaling complex tasks

When a task is too large or complex for a single agent, multi-agent orchestration lets a lead agent break it into pieces and delegate each one to a specialist.

Each specialist has its own model, prompt, and tools. Specialists work in parallel on a shared filesystem and contribute to the lead agent’s overall context. The lead agent coordinates, merges results, and handles the overall task flow.

Practical example

A lead agent receives the instruction: “Audit this website’s SEO and create a report.” It delegates:

Specialist 1: Analyze schema markup and structured data
Specialist 2: Check heading structure and content hierarchy
Specialist 3: Evaluate image optimization and alt text
Specialist 4: Test LLM crawler accessibility

All four specialists work simultaneously. The lead agent collects their findings and produces a unified report. What would take a single agent 30 minutes takes four specialists 8 minutes working in parallel.

Multi-agent orchestration is available in public beta as part of Managed Agents.

Claude Managed Agents with Dreaming multi-agent orchestration and Outcomes — Photo by AISEOShift on AISEOShift

Claude Finance: 10 pre-built agents for financial services

Anthropic shipped Claude Finance, a package of 10 pre-built agents targeting financial services workflows:

Pitch Builder, generates investment pitch decks from data inputs
Meeting Preparer, briefs analysts before client meetings with relevant context
Market Researcher, monitors market developments and surfaces relevant insights
Evaluation Reviewer, reviews valuation models and flags inconsistencies
Month-End Closer, automates reconciliation and reporting workflows
Plus five additional specialized financial agents

Claude Finance is designed for regulated environments with enterprise-grade compliance features. For financial services firms, these agents replace hours of manual research and document preparation with AI-powered workflows that maintain audit trails.

Claude for Small Business: AI in QuickBooks, PayPal, and HubSpot

On May 13, Anthropic launched Claude for Small Business, bringing AI agents into the tools small businesses already use daily.

15 ready-to-run agentic workflows

The launch included 15 pre-built workflows covering:

Payroll planning and processing
Month-end close and financial reporting
Invoice chasing and accounts receivable
Lead triage and qualification
Contract review and summarization
Campaign creation for marketing
Cash-flow monitoring and alerts
Business pulse reporting with weekly summaries

Tool integrations

Connectors ship for QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace, Microsoft 365, and Slack. Claude works inside these tools, it reads your QuickBooks data, drafts invoices, prepares reports, and handles follow-ups.

The critical design principle: Claude asks for approval before anything sends, posts, or pays. It is designed as a drafting and preparation layer, not an autonomous spending agent.

Pricing and availability

Claude for Small Business is available through Claude Team ($25/user/month) or Enterprise plans. The workflows and connectors are included free, partner tool costs (QuickBooks, HubSpot, etc.) remain unchanged.

Anthropic also launched:

Free AI Fluency for Small Business course built with PayPal
10-city US tour starting in Chicago with free half-day workshops
Free month of Claude Max for every workshop attendee

This is Anthropic’s most aggressive go-to-market move. By embedding Claude into the daily workflows of small businesses, Anthropic creates sticky usage that goes beyond the chatbot interface. When Claude handles your invoicing, you do not switch to a competitor.

Claude Security: vulnerability scanning for enterprises

Claude Security launched in public beta for Enterprise customers. It scans code repositories for security vulnerabilities and proposes fixes using Opus 4.7.

This connects directly to Claude Mythos, Anthropic’s cybersecurity-focused model that demonstrated unprecedented vulnerability detection capabilities. Claude Security brings a production-ready version of that capability to enterprise code reviews.

Claude Code desktop app redesign

The Claude Code desktop application received a complete overhaul:

Session sidebar for managing multiple parallel coding sessions
Drag-and-drop workspace for organizing files and projects
Integrated terminal and file editor, no need to switch to a separate editor
Faster diffs with expanded previews
SSH support on Mac for remote development
Performance and reliability improvements across the board

The redesign reflects how developers actually use Claude Code in 2026: running multiple autonomous sessions in parallel, switching between projects, and monitoring long-running tasks. The new sidebar makes it possible to manage 5 to 10 simultaneous coding agents without losing track of what each one is doing.

Claude Design expansion

Claude Design, launched in April, received expanded capabilities in May:

More template options for presentations, wireframes, and social assets
Improved export to Canva, PDF, PPTX, and URL formats
Better design system awareness for brand consistency
Higher-quality visual output powered by continued Opus 4.7 vision improvements

Claude Design is available for Pro, Max, Team, and Enterprise subscribers.

What this means for AI visibility and SEO

Several Code with Claude announcements directly impact AI visibility strategy:

1. Doubled limits mean more AI browsing. With 2x token limits and no peak throttling, Claude users will run more searches, ask more questions, and consume more web content. The volume of AI-mediated content discovery is increasing, making AI visibility optimization more valuable.

2. Multi-agent orchestration changes competitive research. Teams can now run parallel AI agents that audit competitor websites, analyze content gaps, and generate reports, all simultaneously. The speed of competitive intelligence is accelerating.

3. Dreaming creates persistent AI expertise. As Claude agents accumulate knowledge about specific industries and topics through Dreaming, they develop preferences for certain sources. Content that gets cited by Claude agents early builds compounding advantage as those agents’ memories persist.

4. Claude for Small Business expands the AI user base. When small businesses use Claude for invoicing, marketing, and reporting, they also start using it for research, including searching for vendors, products, and services. Being visible to these AI-mediated searches becomes a customer acquisition channel.

Timeline of May 2026 announcements

Date	Announcement
May 6	Code with Claude conference (San Francisco)
May 6	SpaceX Colossus 1 compute partnership
May 6	Doubled limits across all paid plans
May 6	Dreaming, Outcomes, Multi-agent orchestration
May 6	Claude Finance (10 pre-built agents)
May 6	Claude Security (public beta)
May 7	Claude Code desktop app redesign
May 13	Claude for Small Business launch
May 19	Code with Claude London
May 20	Claude Design expanded features
June 10	Code with Claude Tokyo (upcoming)

Frequently asked questions

What is Code with Claude 2026?

Code with Claude is Anthropic’s annual developer conference. The 2026 edition was held in San Francisco on May 6, with follow-up events in London (May 19) and Tokyo (June 10). Major announcements included doubled usage limits, Dreaming for self-improving agents, multi-agent orchestration, Claude Finance, Claude for Small Business, and a compute partnership with SpaceX.

Did Claude really double its limits?

Yes. Pro, Max, Team, and Enterprise plans all received 2x their previous token limits. Peak-hour throttling was removed for Pro and Max accounts. API limits saw increases of 900 to 1,500 percent depending on tier. This was enabled by Anthropic’s deal with SpaceX for the Colossus 1 data center (220,000+ Nvidia GPUs).

What is Claude Dreaming?

Dreaming is a new capability where Claude agents review their past sessions, extract patterns, and curate a memory store. Future agents accessing the same codebase or project can reference these notes, learning from previous sessions’ mistakes and successes. Harvey reported a 6x jump in task completion rates after enabling Dreaming.

What is Claude for Small Business?

Claude for Small Business launched May 13, 2026 with 15 pre-built agentic workflows and connectors to QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace, Microsoft 365, and Slack. Workflows handle payroll, invoicing, lead triage, contract review, marketing campaigns, and financial reporting. Available on Team ($25/user/month) and Enterprise plans.

How does multi-agent orchestration work?

A lead agent breaks complex tasks into pieces and delegates each one to specialist agents. Each specialist has its own model, prompt, and tools, and works in parallel on a shared filesystem. The lead agent coordinates results. This is useful for large-scale tasks like website audits, codebase refactoring, and multi-document analysis.

Is Claude competing with Salesforce and HubSpot now?

Not directly. Claude for Small Business integrates with these platforms rather than replacing them. Claude connects to HubSpot for lead management and to QuickBooks for accounting, it does not offer its own CRM or accounting features. The positioning is “AI layer on top of your existing tools,” not a replacement.