Missed our SF Code w/ Claude developer conference? Keynotes and breakout sessions are now on YouTube.

This week in San Francisco, we hosted Code w/ Claude, our annual developer conference. The event brought together developers, engineers, and founders for two days of keynotes, breakout sessions, and workshops with the teams building Claude.

From prompting and model selection to architecting skills and scaling AI-native engineering teams, every session circled the same shift: the distance between an idea and production software is narrowing, and the teams getting the most leverage are designing for the AI exponential—not reacting to it.

We demonstrated this through live coding sessions, customer deep-dives, and hands-on tutorials highlighting what this looks like today.

What we announced

Announced at the conference, we doubled rate limits on Claude Code and raised API limits for Claude Opus so developers, startups, and enterprises can build more reliably at scale. Both changes are now live.

We also introduced new capabilities to Claude Managed Agents on the Claude Platform aimed at helping teams build and deploy cloud-hosted agents at scale. Four new features are now available to all developers:

• Dreaming. A scheduled process that reviews past agent sessions, surfaces patterns, and curates memory, so agents improve between runs. Recurring mistakes, shared workflows, and team preferences get pulled into a more useful memory store.
• Multiagent orchestration. A lead agent can delegate to specialist subagents working in parallel on a shared filesystem, each with its own model, prompt, and tools. The whole flow is traceable in the Claude Console.
• Outcomes. Developers define a rubric for what a good output looks like. A separate grader evaluates each result in its own context window and sends the agent back to revise until it meets the bar. On our internal benchmarks, outcomes lifted task success by up to 10 points on the hardest problems.
• Webhooks: Once you’ve defined an outcome, you can let the agent run, and get notified by a webhook when it's done.

In case you missed it

If you missed the livestream, check out our keynote and breakout session recordings, here.

Our talks go behind the scenes of building Claude with Anthropic teams and share how customers like Asana, Cursor, GitHub, Replit, and Vercel about how they’re designing production-ready agents and pushing the boundaries of agentic development.

We’ll be taking Code w/ Claude to London (May 20-21) and Tokyo (June 5-6). All Day 1 keynotes and breakout sessions will be streamed live.

Stay tuned for technical tutorials, guides, and customer stories inspired by our talks.

Original source

We’re releasing 20+ new MCP connectors that link Claude to the software that the legal industry runs on and 12 new plugins tailored to specific legal work and practice areas.

Earlier this year we released our first legal plugin, and in the months since, legal professionals have become the most engaged Claude Cowork users of any knowledge-work function. We’re now building on that with a much larger set of tools.

Legal work runs on a specific technology stack: contract lifecycle systems, research platforms, document management, e-discovery, data rooms, firm-specific precedents, and much more. Claude now connects to all of it through several key building blocks. First, MCP connectors bring your legal work (the documents, communications, and records tied to specific matters) into Claude. Secondly, practice-area plugins package the tasks that lawyers run most often. And finally, because both are built on open protocols, firms and in-house teams can customize Claude to match the way they actually practice.

Today we’re introducing 20+ new MCP connectors that link Claude to the software the legal industry already relies on, and 12 new plugins tailored to specific legal work and practice areas. And finally, we're partnering with the Free Law Project, the Justice Technology Association, and others working to put legal help within reach of people who can’t currently access it.

Claude works where legal teams work

Claude meets legal teams where they are, working directly inside Microsoft Word, Outlook, Excel, and PowerPoint while carrying context across all four apps. A redline finished in Word doesn’t need to be re-explained when it becomes a cover note in Outlook, a closing checklist in Excel, or a board summary in PowerPoint.

In Word, Claude skills—reusable instructions that encode a team's playbooks and standards—handle drafting, redlining, and clause-by-clause comparisons against those playbooks, tracking every change and explaining the reasoning behind it. They also take on the rote work in between every turn: scrubbing internal comments before a draft goes to the counterparty, running a final formatting check on an execution copy, and pulling fallback language from approved playbooks.

In Outlook, Claude triages incoming matter work: flagging contract requests, drafting responses and cover notes, and scheduling follow-ups so nothing pile ups.

In Claude Cowork, the same connectors and plugins are available for work that spans many documents: triaging a batch of contracts, clearing a product feature for launch, and drafting a note on regulatory developments for the board. Scheduled tasks can automate recurring work such as weekly regulatory update sweeps or intake triage.

And with Projects, matter teams get a persistent workspace where precedents and prior drafts are retained across every conversation.

New connectors across the legal stack

New MCP connectors let Claude interact with the systems that legal teams already rely on.

Contract lifecycle and drafting:

• Definely, which gives live, deterministic access to contract structure for review: resolve definitions, validate cross-references, map dependencies, and run structural diffs to see how edits propagate across an agreement;
• Docusign, which connects Claude to your agreement data so you can quickly surface key terms like renewal dates and obligations, and orchestrate agreement workflows across the contract lifecycle, from drafting through signature and post signature management;
• Ironclad, which lets Claude access your contract repository and workflows and ask questions about contracts in plain language, with results automatically scoped to each user's permissions.

Deal rooms and transaction documents:

• Box, which connects Claude to content stored in Box to search and access files, query documents, create or update content, and extract metadata fields, while enforcing existing Box security and access policies;
• Datasite, which connects Claude to your Datasite virtual data room—the secure workspace where thousands of M&A deals are facilitated annually—to set up folder structures, invite users, search documents, track buyer Q&A, and audit data room readiness.

Document management:

• iManage, a knowledge work platform, gives Claude permission-bound, auditable access to governed iManage content, including matter history, documents, and institutional knowledge, eliminating the need for bulk exports or custom integrations;
• NetDocuments, which lets Claude search and retrieve documents from your NetDocuments repository and draft new documents based on your precedents, with full respect for your organization's permissions and governance policies.

Expert networks and skills:

• Lawve AI, which offers a curated library of legal AI skills written by practicing lawyers, in-house counsel, and legal technologists, searchable from inside Claude;
• The L Suite, which offers two MCP connectors from their leading in-house counsel community: (1) Lloyd, which allows The L Suite members to connect Claude with the Braintrust member platform; and (2) TopCounsel, which helps any in-house counsel find the right outside counsel for a specific matter based on The L Suite's proprietary dataset and ranking algorithm.

E-discovery and review:

• Consilio, which puts a client’s own live matters and Consilio’s Aurora Legal AI at Claude's fingertips, with every response scoped to what the user is already entitled to see;
• Everlaw, which provides a litigation platform, lets Claude search, organize, and retrieve documents from Everlaw projects using metadata, keywords, and document types, with direct review links;
• Relativity, which lets Claude stand up matters, shape workspace schema, govern access, and analyze usage in its AI platform for legal data intelligence, RelativityOne.

Fiduciary-grade workflows:

• Thomson Reuters, which connects Claude to CoCounsel Legal, a fiduciary-grade system for end-to-end drafting, research, review, and validation across all major practice areas, serving as an AI assistant for high-stakes legal work, grounded in Westlaw primary law, Practical Law guidance, KeyCite, and your own documents, with transparent and verifiable outputs.

Legal research and case law:

• Legal Data Hunter, which gives Claude access to the world's fastest growing legal corpus: 31M+ documents from 160+ jurisdictions, including EU consolidated law, case law from supreme and constitutional courts, and official doctrine;
• Midpage, which connects Claude to a database of case law for complex legal research, opinion review, and work product, with everything hyperlinked to real sources for easy verification;
• Trellis, which gives Claude direct access to the largest state trial-court dataset in the US, including dockets, rulings, verdicts, and filings, for judge and opposing-counsel analytics and motion drafting.

Legal AI assistants:

• Harvey, which brings Harvey's legal intelligence into Claude, supporting general legal inquiries, analysis over Vault projects, and research questions for select knowledge sources;
• Solve Intelligence, which connects Claude to patent and non-patent literature, legal texts, SEP technical standards, and the open web for prior-art search, claim mapping, and patent drafting.

Public service

• BoardWise, which guides licensed professionals facing state board matters: helping them understand deadlines, navigate their situation, and draft structured response letters tailored to their jurisdiction;
• Courtroom5, which provides legal guidance to the roughly 80% of civil litigants who appear in court without an attorney, with jurisdiction-aware case intake, deadline calculation, and next-step guidance across all 50 states;
• Descrybe, which gives Claude legal-research tools for working with primary law: search cases by concept or citation, check treatment status, find citing authorities, and verify quoted language;
• Free Law Project, which connects Claude to CourtListener's millions of US court opinions, PACER dockets, judge profiles, oral arguments, and citation data.

Practice-area plugins

Legal work looks different depending on the seat you're in. We're releasing 12 practice-area plugins (download them from the Legal Marketplace, here), each built around a specific legal role. Every plugin starts with a short setup interview that learns your practice: your playbook, your escalation chain, your risk calibration, your house style, so Claude’s answers are not generic but rather tailored for your team. These include:

• Commercial Legal reviews vendor agreements and NDAs against your playbooks and routes escalations with a plain-language summary for business stakeholders.
• Corporate Legal handles M&A: diligence across the data room, disclosure schedules, board consents, and the closing checklist. It can be configured for board work, public-company governance, or entity compliance.
• Employment Legal covers hires, terminations, worker classification, leave deadlines, and investigations, and drafts policies with state-specific rules built in.
• Privacy Legal reviews DPAs against your playbook, triages PIAs and DPIAs, prepares DSAR responses within statutory timelines, and flags gaps between written policy and actual practice.
• Product Legal runs launch reviews against your internal framework, checks marketing claims for substantiation, and answers risk questions from teams across the business.
• Regulatory Legal monitors regulatory developments, filters them to your materiality threshold, compares new rules against your policy library, and tracks gaps and comment deadlines.
• AI Governance Legal triages AI use cases against your governance tiers, runs impact assessments, reviews vendor AI terms, and can draft a starting AI policy.
• IP Legal conducts trademark clearance and freedom-to-operate triage, drafts and responds to cease-and-desist letters, handles DMCA takedowns and open-source compliance, and screens invention disclosures.
• Litigation Legal manages matter intake and portfolio tracking, legal holds, demand letters, subpoena triage, chronologies, deposition preparation, privilege logs, and brief drafting.
• Law Student provides Socratic drilling, case briefs, IRAC grading, and bar preparation with jurisdiction-specific distinctions.
• Legal Clinic manages client intake, deadline tracking, case memos, and the supervisor review queue.
• Legal Builder Hub finds and installs community-built legal skills from public registries, running a security review, license check, and freshness check on every install and update.

Each agent template can be installed in Cowork or Claude Code with a click, and produces outputs that match institutional drafting standards. A subset of these (Commercial Legal, Corporate Legal, Litigation Legal, Product Legal, Litigation Legal) are also available as cookbooks that can be deployed as Managed Agents in the Claude Platform for programmatic use. Teams can layer on their own precedents and playbooks to customize the skills.

Every legal organization works differently, and no single set of plugins can cover every practice. The plugin and skill ecosystem are open protocols, and early contributors including Box, Legal Quants, Lawve AI, and Thomson Reuters have already shipped skills, plugins, and style conventions of their own. Any partner can submit connectors and skills through the Directory.

Democratizing access to legal services with AI

Legal services are out of reach for many people and small businesses, and the gap is widening. We’re working with the Free Law Project, Justice Technology Association, and other legal aid and Public Service organizations to help make legal services more affordable and available.

Qualifying legal aid clinics, public defenders, and nonprofit legal services organizations can gain access to significantly discounted pricing through the Claude for Nonprofits program. Free and low-cost tools from BoardWise, Courtroom5, Descrybe, and Free Law Project are available to Claude users via MCP connectors as well.

"Most people don't know they have legal rights until it's too late to use them. Claude can now meet them where they are — in the moment they're scared and searching for answers."

• Sonja Ebron, CEO & Co-Founder, Courtroom5

Trusted across the legal industry

Firms and in-house teams have moved from testing Claude to running their practice on it — and the legal tools they rely on are increasingly built on Claude too. At our Briefing: Enterprise Agents in February, Thomson Reuters showcased CoCounsel rebuilt on the Claude Agent SDK; with today's release, that integration runs both ways. Harvey, Solve Intelligence, and others below are doing the same.

These updates build on Claude Opus 4.7, our most capable publicly available model for legal reasoning and long-document work.

Here’s what legal teams and ecosystem partners have told us about working with Claude:

[Multiple testimonials from legal industry leaders praising Claude's integration, capabilities, and impact on legal workflows and AI transformation.]

Getting started

The new connectors and practice-area plugins are open source and available in Claude Cowork. Enterprise admins can enable them in your workspace settings. Learn more here or contact our sales team.

Register for our launch webinar to see the new features in action – we'll share live product walkthroughs, and cover how law firms and in-house teams can get the most out of Claude and the connected ecosystem. Builders can also explore community-maintained skills via Legal Quants and Lawvable.

For legal aid and access-to-justice organizations who are interested in partnering, get in touch via our Nonprofits program.

Original source

How it works

Today we're introducing agent view in Claude Code: one place to manage all your Claude Code sessions.

When running agents in parallel before, you've probably had to manage multiple terminal tabs, a tmux grid, and an overloaded mental ledger of what you need to tackle next.

With agent view in Claude Code, you can kick off new agents, send them to the background, and jump in only when Claude needs you. See at a glance which agents are waiting on you, which are still working, and which are done, so you can easily steer many all at once.

Agent view improves visualizing and interacting with your Claude Code sessions in the CLI.

See everything at once

Press the left arrow from any session or run claude agents from the terminal to open agent view. Each row shows the session, whether it needs your input, the contents of its last response, and when you last interacted with it.

Peek and reply without leaving

Select a session to peek at the last turn. If a session is waiting on a decision, answer inline and the session picks back up. Press enter to attach directly to sessions where you want to explore the full transcript.

Background anything

Lastly, users can take any existing session and add it to agent view using /bg or skip the foreground entirely using claude --bg [task] to launch a fresh session.

How developers are using agent view

A few patterns we have seen from early users:

• Scaling the number of concurrent sessions: Dispatch several ideas at once, each optionally paired with a skill, and return to a list of pull requests ready for review.
• Manage long running agents: PR babysitters, dashboard updaters, and other looping jobs show their next run time right in the list.
• Navigate between separate sessions: When you’re in the middle of a session, press the left arrow, start a related task or quick codebase question, then arrow right back into what you were doing. Peek shows the answer when it lands.
• See what shipped: Status indicators on each row plus the title in peek make it easy to scan which sessions produced a PR.

Getting started

Agent view is available today as a Research Preview on Pro, Max, Team, Enterprise, and Claude API plans. Opt-in by running claude agents. Usual rate limits apply. See the docs for more information.

Original source

The Claude Platform on AWS is now generally available, offering a new way for AWS customers to access the full set of Claude Platform features with AWS authentication, billing, and commitment retirement. Claude also remains available on Amazon Bedrock, where AWS is the data processor.

Starting today, Claude Platform on AWS customers can deploy agents at scale with Claude Managed Agents and build with tools like code execution, skills, the advisor strategy, and more.

Access the complete Claude Platform via AWS

The Claude Platform on AWS brings the full set of Claude API features to AWS customers for the first time, with all new features and betas shipping the same day they go live on the native Claude API.

Authentication runs through AWS IAM, audit logging through CloudTrail, and billing through a single AWS invoice that fully retires against existing commitments. Customers use their existing AWS credentials and IAM policies, so teams stay within the tools and permissions they already manage.

Claude Platform on AWS will be available in most AWS commercial regions and support global and U.S. inference geographies.

What's included

The Claude Platform on AWS includes native platform features, like:

• Claude Managed Agents (beta) to build and deploy agents at scale
• Advisor strategy (beta) to give agents an intelligence boost by consulting an advisor model
• Web search and web fetch to augment Claude’s knowledge with current, real-world data from across the web
• Code execution to run Python code, create visualizations, and analyze data directly within API calls
• Files API (beta) for uploading and referencing documents across conversations
• Skills (beta) to teach Claude best practices so it delivers consistent results
• MCP connector (beta) to connect Claude to any remote MCP server without writing client code
• Prompt caching for reducing costs and latency on repeated context
• Citations for grounding responses in source documents
• Batch processing for high-volume, asynchronous workloads

Claude Platform on AWS customers also get access to the Claude Console, Anthropic's development environment for building and testing with Claude. The Console includes a prompt improver, prompt generator, and evaluation tools.

Claude Opus 4.7, Sonnet 4.6, and Haiku 4.5 are available, with new models shipping on the Claude Platform on AWS as they launch.

Choosing the right path for developers

Both the Claude Platform on AWS and Claude on Amazon Bedrock enable AWS customers to build on Claude models. The difference is in who operates the service and which features are available.

The Claude Platform on AWS is a first of its kind offering for Anthropic, giving you all native Claude API features from day one. Anthropic operates the service and data is processed outside the AWS boundary. This is a good option for companies that want the full Claude Platform experience.

Claude on Amazon Bedrock keeps AWS as the data processor and operates within the AWS boundary. This is a good fit for companies that have strict regional data residency requirements or need their data processed exclusively within AWS's infrastructure.

Getting started

The Claude Platform on AWS is available today. To get started, visit the Claude Platform on AWS or explore the documentation.

If you have an existing Bedrock private offer, please contact your Anthropic or AWS account executive before getting started with Claude Platform on AWS to ensure your discounts are applied correctly. Discounts cannot be applied retroactively to usage incurred before a Claude Platform private offer is accepted.

Original source

Starting today, Claude for Excel, PowerPoint, and Word are generally available, and Claude for Outlook is now in public beta for all paid plans. As Claude moves between tasks in your Microsoft apps, it keeps the full context of your conversation.

One conversation across four apps

Most work doesn't live in one application, so Claude moves between them as you do. Triage an email in Outlook, open the attachment in Word to draft a memo in your team's template, build the supporting analysis in Excel, and turn it into a deck in PowerPoint — all without re-explaining what you're working on. Adjust an assumption in Excel and the chart in PowerPoint and the number in your Word memo automatically update, too.

Claude works across open files, so keep the spreadsheet, deck, and memo open side by side and changes will flow between them.

Conversations persist with each file, so you can close the sidebar, reopen it the next day, and pick up where you left off using your keyboard or your voice.

Claude for Outlook

Claude for Outlook brings Claude into your inbox. Ask Claude to triage your inbox and it sorts messages by what needs your response, what it can draft for you, and what's noise. Replies land as drafts in Outlook's compose pane with recipients, subject, and body filled in. Calendar invites check attendee availability and open in Outlook's native event form.

Open an attachment in Word or Excel and Claude already knows what the sender asked for. You review every reply and calendar invite before it goes out, and nothing goes out until you click send.

Deploying Claude for Microsoft 365 across your organization

Claude for Excel, PowerPoint, and Word are now generally available, with the controls IT admins and organizations need. One AppSource listing covers Excel, PowerPoint, and Word, and a separate listing adds Outlook in beta. Admins can deploy both from the Microsoft admin center.

Enterprise admins can configure OpenTelemetry to stream prompts, tool calls, and document references to their own collector, so security teams see exactly what Claude does across every app. The Analytics API breaks out activity per user, per app, per day.

Organizations can access all four add-ins with a Claude account, or route traffic through an existing LLM gateway to Claude models running on Amazon Bedrock, Google Cloud’s Vertex AI, or Microsoft Foundry.

Microsoft 365 Copilot customers can also work with Claude AI models directly within Excel and PowerPoint.

How organizations use Claude for Microsoft 365

“We can instruct Claude to build a style guide from an executive's prior written communications so our EAs can draft their emails in their voice. That's high-leverage work the team would never have time to set up by hand.”
Ben Letalik, Sr. Director, Digital Transformation & Innovation

“We can instruct Claude to build a style guide from an executive's prior written communications so our EAs can draft their emails in their voice. That's high-leverage work the team would never have time to set up by hand.”
Rajeev Sethi, GVP Enterprise Technologies

“Our investment professionals live in data and analytical models, and Claude for Excel meets them there. Analysts are using it to build and update coverage models, separate signal from noise, and pressure-test their work — all with a step-change in efficiency.”
Atte Lahtiranta, Head of Core Engineering

“Claude in Excel allows my teams to build an initial version of complex models faster, enabling them to focus on refining the model, pressure testing inputs and assumption, and exploring more scenarios and trade-offs. This helps us have richer and deeper discussions with our clients earlier.”
Gene Rapoport, Head of Private Equity AI Practice

“Claude enables a seamless transition from rough ideas to polished, branded deliverables across presentation, document and spreadsheet applications without disrupting the workflow. That end-to-end consistency has been a significant time-saver, especially when quality, formatting, and speed all matter.”
Vivek Kulkarni, US AI Transformation Leader

Getting started

All Mac and Windows users on paid plans can access Claude for Microsoft 365. Claude for Outlook is available in beta on all paid plans. Admins can deploy these add-ins from Microsoft AppSource through the Microsoft admin center.

Original source

Build self-improving agents with dreaming

Today we're launching dreaming in Claude Managed Agents as a research preview. Dreaming extends memory by reviewing past sessions to find patterns and help agents self-improve. We're also making outcomes, multiagent orchestration, and webhooks available to developers building with Managed Agents. Together, these updates make agents more capable at handling complex tasks with minimal steering.

Dreaming is a scheduled process that reviews your agent sessions and memory stores, extracts patterns, and curates memories so your agents improve over time. You decide how much control you want: dreaming can update memory automatically, or you can review changes before they land.

Dreaming surfaces patterns that a single agent can’t see on its own, including recurring mistakes, workflows that agents converge on, and preferences shared across a team. It also restructures memory so it stays high-signal as it evolves. This is especially useful for long-running work and multiagent orchestration.

Together, memory and dreaming form a robust memory system for self-improving agents. Memory lets each agent capture what it learns as it works. Dreaming refines that memory between sessions, pulling shared learnings across agents and keeping it up-to-date.

Dreaming is available in Managed Agents on the Claude Platform; developers can request access here.

Deliver better outcomes

With outcomes, you write a rubric describing what success looks like and the agent works toward it. A separate grader evaluates the output against your criteria in its own context window, so it isn't influenced by the agent's reasoning. When something isn't right, the grader pinpoints what needs to change and the agent takes another pass.

Agents do their best work when they know what "good" looks like. For example, a structural framework, a presentation standard, or a set of requirements that need to be met. With outcomes, agents can check their work against that bar and self-correct until the output is good enough, without a human needing to review each attempt.

Outcomes is particularly useful for tasks that require attention to detail and exhaustive coverage. It also works for subjective quality, like whether copy matches a brand voice or a design follows visual guidelines. In testing, outcomes improved task success by up to 10 points over a standard prompting loop, with the largest gains on the hardest problems. Outcomes also improved file generation quality, with +8.4% task success on docx and +10.1% on pptx in our internal benchmarks.

You can also now define an outcome, let the agent run, and get notified by a webhook when it's done.

Handle complex tasks with multiple agents

When there is too much work for a single agent to do well, multiagent orchestration lets a lead agent break the job into pieces and delegate each one to a specialist with its own model, prompt, and tools. For example, a lead agent can run an investigation while subagents fan out through deploy history, error logs, metrics, and support tickets.

These specialists work in parallel on a shared filesystem and contribute to the lead agent's overall context. The lead agent can check back in with other agents mid-workflow because events are persistent and every agent remembers what it's done. You can also trace every step in the Claude Console: which agent did what, in what order, and why, giving you full visibility into how your task was delegated and executed.

What teams are building

Teams are using dreaming, outcomes, and multiagent orchestration to ship agents that verify their own work, learn across sessions, and parallelize complex jobs:

• Harvey uses Managed Agents to coordinate complex legal work like long-form drafting and document creation. With dreaming, their agents remember what they learned between sessions, including filetype workarounds and tool-specific patterns. Completion rates went up ~6x in their tests.
• Netflix's platform team built an analysis agent that processes logs from hundreds of builds across different sources. With changes that affect thousands of applications, what matters is finding the issues that recur across many of them. Multiagent orchestration lets the agent analyze batches in parallel and surface only the patterns worth acting on.
• Spiral by Every is using multiagent orchestration and outcomes to power the writing agent behind their new API and CLI. The lead agent runs on Haiku: it fields incoming requests, poses quick follow-up questions when needed, then delegates the drafting to subagents running on Opus. When a user asks for multiple drafts, the subagents run in parallel. Writing quality is Spiral's core value, so they use outcomes to enforce it. Each draft is scored against a rubric of Every's editorial principles and the user's voice, both pulled from memory. Only drafts that clear the bar are returned.
• Wisedocs built a document quality check agent on Managed Agents, using outcomes to grade each review against their internal guidelines. Reviews now run 50% faster, while staying aligned with their team's standards.

Getting started

Dreaming is available in research preview, outcomes, multiagent orchestration, and memory are available in public beta as part of Managed Agents. To get started with dreaming, request access here. Explore our documentation to learn more or visit the Claude Console to deploy your first agent.

Original source

Claude Security is now in public beta for Claude Enterprise customers. Scan code for vulnerabilities and generate proposed fixes with Opus 4.7, on the Claude Platform, or through technology and services partners building with Claude.

Claude Security is now available in public beta to Claude Enterprise customers.

AI cybersecurity capabilities are advancing fast. Today’s models are already highly effective at finding flaws in software code; the next generation will be more capable still, and will be particularly effective at autonomously exploiting these flaws. Now is the time for organizations to act to improve their security, preparing for a world in which working software exploits are much easier to discover.

Recently, we made Claude Mythos Preview—which can match or surpass even elite human experts at both finding and exploiting software vulnerabilities—available to a number of partners as part of Project Glasswing.

But our cybersecurity efforts go beyond Glasswing: with Claude Security, a much wider set of organizations can put our most powerful generally-available model, Claude Opus 4.7, to work across their codebases. Opus 4.7 is among the strongest models available for finding and patching software vulnerabilities, and for discovering complex, context-dependent issues that might otherwise be missed.

Claude Security—previously known as Claude Code Security—has already been tested by hundreds of organizations of all sizes in limited research preview, helping teams scan their codebases for vulnerabilities and generate targeted patches. Their feedback has shaped today’s release, which makes Claude Security available to all Enterprise customers. It comes with scheduled and targeted scans, easier integration with audit systems, and improved tracking of triaged findings. No API integration or custom agent build is required: if your organization uses Claude, you can start scanning today.

Opus 4.7’s capabilities are also being brought to cyber defenders through Claude’s integration into software tools that many enterprises already use. Our technology partners, including CrowdStrike, Microsoft Security, Palo Alto Networks, SentinelOne, TrendAI, and Wiz are embedding Opus 4.7 into their tools; in addition, services partners like Accenture, BCG, Deloitte, Infosys and PwC are now helping organizations deploy Claude-integrated security solutions.

We are entering a pivotal time for cybersecurity. AI is compressing the timeline between vulnerability discovery and exploitation. We believe the right response is to make sure defenders have access to frontier capabilities in the ways most accessible to them, through Claude directly and through our partners.

How Claude Security works

Claude Security can be accessed directly from the Claude.ai sidebar, or at claude.ai/security. To begin, select one of your repositories (or scope to a specific directory or branch), then start a scan.

While scanning, Claude reasons about code much like a security researcher. Rather than finding vulnerabilities by searching for known patterns, Claude seeks to understand how components interact across files and modules, traces data flows, and reads the source code.

Once complete, Claude provides a detailed explanation of each of its findings, including its confidence that the vulnerability is real, how severe it is, its likely impact, and how it can be reproduced. It also generates instructions for a targeted patch, which users can open in Claude Code on the Web to work through the fix in context.

What we've learned since our initial preview

Over the past two months, we’ve refined Claude Security in line with what we learned from its use in production across hundreds of enterprises. Specifically, we’ve seen that:

Detection quality is paramount.

Teams have told us that high-confidence findings are what really accelerate security work. Claude Security's multi-stage validation pipeline independently examines each finding before it reaches an analyst, which drives down false positives, and Claude attaches a confidence rating to every result. This means that the signal that reaches the team is worth acting on.

Time from scan to fix is the metric that matters.

Early users pointed to this consistently, with several teams going from scan to applied patch in a single sitting, instead of days of back-and-forth between security and engineering teams.

Teams want ongoing coverage, not one-off audits.

We've added the option to schedule scans, so teams can set a regular cadence around reviewing and acting on findings.

With this release, we've also added the ability to target a scan at a particular directory within a repository, dismiss findings with documented reasons (so that future reviewers can trust prior triage decisions), export findings as CSV or Markdown for existing tracking and audit systems, and send scan results to Slack, Jira, or other tools via webhooks.

Here, organizations who’ve used Claude Security describe their experience:

“We are adapting our proactive security efforts through our Anthropic partnership. Claude Security helps us accelerate how we generate and secure new code at the scale and speed of DoorDash— it surfaces deep vulnerabilities accurately, and pipes findings right into our workflows so engineers can act on them in context.”
Suha Can, Vice President and Chief Security Officer

“Claude Security surfaced novel, high-quality findings during our early testing of the research preview that helped us identify and address potential security issues before they could affect our environment or our customers. We see strong potential as we expand its use.”
Krzysztof Katowicz-Kowalewski, Staff Product Security Engineer

"Claude Security grasps the actual business logic behind our code. Our security team can now go from scan to fixes in a few clicks within our trusted tooling."
Greg Janowiak, Information Security Officer

"The scan quality is why we're working to plug Claude Security directly into our vulnerability management program—so real issues reach engineering faster, with less triage overhead in between."
Chiara La Valle, Head of Security

"Given the increasing pace of vulnerability discovery, the strongest signal for us is how quickly a finding turns into a PR we can actually merge, not a ticket. We've used patches built with Claude Security to close real vulnerabilities in minutes, not days."
Matt Aromatorio, Head of Security

We're still in the early days of AI-powered security. As our models improve and as we learn from the teams using Claude Security in production, we'll continue to expand what these tools can do.

Meeting defenders where they work

Claude Security is one part of our broader push to put frontier capabilities in defenders' hands. Much of its reach comes through the platforms and services partners teams rely on today.

CrowdStrike, Microsoft Security, Palo Alto Networks, SentinelOne, TrendAI, and Wiz are integrating Opus 4.7’s capabilities into the security platforms that enterprises already run on today.

Accenture, BCG, Deloitte, Infosys, and PwC are working alongside enterprise security organizations to deploy Claude-integrated security solutions for vulnerability management, secure code review, and incident response programs.

Together, this means organizations can adopt these capabilities through whichever path fits how they already operate: directly in Claude Security, embedded in a platform they trust, or with a services team guiding the rollout.

Getting started

Claude Security is available in public beta starting today for Claude Enterprise customers. Access for Claude Team and Max customers is coming soon.

Admins can enable Claude Security in the admin console. For a full walkthrough, see our Getting Started Guide.

Claude Opus 4.7 uses new cyber safeguards that automatically detect and block requests that are suggestive of prohibited or high-risk cybersecurity uses. However, organizations conducting work that may trigger these safeguards can become a part of our Cyber Verification Program, which is part of our effort to make frontier capabilities available to defenders while keeping them out of the wrong hands.

Original source

How people have been using connectors

Since launching in July 2025, the Claude directory has grown to over 200 connectors, spanning popular apps for design, finance, productivity, and health used by millions of people every day. We’ve found that people often connect multiple apps and use them together in a single conversation with Claude. For example, a product manager will pull a query from Amplitude, turn it into a Canva deck, and drop the link into Asana for the team, all without leaving the conversation.

This expansion extends that functionality to more of the tools you use outside of work. Starting today, Claude connects to AllTrails, Audible, Booking.com, Instacart, Intuit Credit Karma, Intuit TurboTax, Resy, Spotify, StubHub, Taskrabbit, Thumbtack, Tripadvisor, Uber, Uber Eats, and Viator, with more on the way.

Connectors dynamically show up in conversations

We’ve also changed how connectors show up in conversations. Claude now suggests the right app for what you’re doing, like finding a reservation, adding to a grocery cart, or identifying a flight. It’s working from what you’ve told it: your preferences, your context, your conversation.

This all happens inside your conversation. Ask Claude to recommend a weekend hike, and AllTrails will surface trails nearby that match your preferences. You can then refine the recommendations by asking for something shorter, more scenic, or dog-friendly without leaving the thread. When more than one connected app could help answer your question, Claude shows them all and lets you choose.

You control what you see and share

Claude is ad-free and will stay that way. There are no paid placements or sponsored answers in conversations with Claude. When two connectors could help, you see both, ranked by what’s most useful to you.

Connecting a service gives Claude access to it on your behalf. Your data from that app isn’t used to train our models, and the app doesn’t see your other conversations with Claude. You can also disconnect it at any time.

Claude suggests connectors and makes recommendations. But you stay in control of its actions: before it books or purchases something on your behalf, it’s designed to check with you first.

Getting started

The more apps you connect, the more Claude can do. Claude suggests relevant connectors as you work, which you can install with a click on desktop or a few taps on mobile. Once connected, the service is available in every conversation.

Connectors are available on all plans, with mobile in beta. See the full list at claude.ai/directory/connectors. If you build a product that would be useful in Claude, submit it to our directory here.

For more information on getting started with connectors, along with security and privacy best practices when connecting data sources to Claude, visit our Help Center.

Original source

Memory on Claude Managed Agents

Memory on Claude Managed Agents is available today in public beta. Your agents can now learn from every session, using an intelligence-optimized memory layer that balances performance with flexibility. Because memories are stored as files, developers can export them, manage them via the API, and keep full control over what agents retain.

Agents that learn across sessions

Managed Agents pairs production infrastructure with a harness tuned for performance. Memory extends that: it’s optimized against internal benchmarks for long-running agents that improve across sessions and share what they've learned with each other.

We've found that agents are most effective with memory when it builds on the tools they already use. Memory on Managed Agents mounts directly onto a filesystem, so Claude can rely on the same bash and code execution capabilities that make it effective at agentic tasks. With filesystem-based memory, our latest models save more comprehensive, well-organized memories and are more discerning about what to remember for a given task.

Portable memories for production-grade agents

Memory is built for enterprise deployments, with scoped permissions, audit logs, and full programmatic control. Stores can be shared across multiple agents with different access scopes. For example, an org-wide store might be read-only, while per-user stores allow reads and writes. Multiple agents can work concurrently against the same store without overwriting each other.

Memories are files that can be exported and independently managed via the API, giving developers full control. All changes are tracked with a detailed audit log, so you can tell which agent and session a memory came from. You can roll back to an earlier version or redact content from history. Updates also surface in the Claude Console as session events, so developers can trace what an agent learned and where it came from.

What teams are building

Teams have been using memory to close feedback loops, speed up verification, and replace custom retrieval infrastructure:

• Netflix agents carry context across sessions, including insights that took multiple turns to uncover and corrections from a human mid-conversation, instead of manually updating prompts and skills.
• Rakuten's task-based long-running agents use memory to learn from every session and avoid repeating past mistakes, cutting first-pass errors by 97%, all within workspace-scoped, observable boundaries.
• Wisedocs built their document verification pipeline on Managed Agents, using cross-session memory to spot and remember recurring document issues, speeding up verification by 30%.
• Ando is building their workplace messaging platform on Managed Agents, capturing how each organization interacts instead of building memory infrastructure themselves.

Getting started

Memory on Managed Agents is now available in public beta on the Claude Platform. Visit the Claude Console or use our new CLI to deploy your first agent with memory. Explore the documentation to learn more.

Original source

The new desktop experience

Today, we're releasing a redesign of the Claude Code desktop app, built to help you run more Claude Code tasks at once.

It includes a new sidebar for managing multiple sessions, a drag-and-drop layout for arranging your workspace, an integrated terminal and file editor, plus performance and quality-of-life improvements.

For many developers, the shape of agentic work has changed. You're not typing one prompt and waiting. You're kicking off a refactor in one repo, a bug fix in another, and a test-writing pass in a third, checking on each as results come in, steering when something drifts, and reviewing diffs before you ship.

The new app is built for how agentic coding actually feels now: many things in flight, and you in the orchestrator seat.

Run sessions in parallel

The new sidebar puts every active and recent session in one place. Kick off work across multiple repos and move between them as results arrive.

You can filter by status, project, or environment, or group the sidebar by project to find and resume sessions faster. When a session's PR merges or closes, it archives itself so the sidebar stays focused on what's live.

When you need to ask a question mid-task, you can open a side chat (⌘ + ; or Ctrl + ;) to branch off a conversation. Side chats pull context from the main thread, but don’t add anything back to the thread, to avoid misdirecting your tasks.

Review and ship without leaving the app

The redesign brings more commonly-used tools into the app, so you can review, tweak, and ship Claude's work without bouncing to your editor:

• Integrated terminal: Run tests or builds alongside your session.
• In-app file editor: Open files, make spot edits directly, and save changes.
• Faster diff viewer: Rebuilt for performance on large changesets.
• Expanded preview: Open HTML files or PDFs in-app, in addition to running local app servers in the preview pane.

Every pane is drag-and-drop. Arrange the terminal, preview, diff viewer, and chat in whatever grid matches how you work.

Fits your stack

The desktop app now has parity with CLI plugins. If your org manages Claude Code plugins centrally, or you've installed your own locally, they work in the desktop app exactly the way they do in your terminal.

You can still run sessions locally or in the cloud. SSH support now extends to Mac alongside Linux, so you can point sessions at remote machines from either platform.

Customize for how you work

Three view modes—Verbose, Normal, and Summary—let you dial the interface from full transparency into Claude's tool calls to just the results. New keyboard shortcuts cover session switching, spawning, and navigation; press ⌘ + / (or Ctrl + /) to see the full list. A new usage button shows both your context window and session usage at a glance.

Under the hood, the app has been rebuilt for reliability and speed, and now streams responses as Claude generates them.

Getting started

The redesigned desktop app is available now for all Claude Code users on Pro, Max, Team, and Enterprise plans, and via the Claude API.

Download the app, or update and restart if you already have it. Explore the documentation to learn more.

Original source

Today, we're introducing routines in Claude Code in research preview.

A routine is a Claude Code automation you configure once — including a prompt, repo, and connectors — and then run on a schedule, from an API call, or in response to an event. Routines run on Claude Code’s web infrastructure, so nothing depends on your laptop being open.

Developers already use Claude Code to automate the software development cycle, but until now, they've managed cron jobs, infrastructure, and additional tooling like MCP servers themselves. Routines ship with access to your repos and your connectors, so you can package up automations and set them to run on a schedule or trigger.

How it works

Scheduled routines

Give Claude Code a prompt and a cadence (hourly, nightly, or weekly) and it runs on that schedule:

Every night at 2am: pull the top bug from Linear, attempt a fix, and open a draft PR.

If you're using /schedule in the CLI, those tasks are now scheduled routines.

API routines

You can also configure routines to be triggered by API calls. Every routine gets its own endpoint and auth token. POST a message, get back a session URL. Wire Claude Code into your alerting, your deploy hooks, your internal tools—anywhere you can make an HTTP request:

Read the alert payload, find the owning service, and post a triage summary to #oncall with a proposed first step.

Webhook routines, starting with GitHub

Subscribe a routine to automatically kick off in response to GitHub repository events. Claude will create a new session for every PR matching your filters and run your routine.

Please flag PRs that touch the /auth-provider module. Any changes to this module need to be summarized and posted to #auth-changes.

Claude opens one session per PR and will continue to feed updates from that PR to the session, so it can address follow-ups like comments and CI failures.

We plan to expand webhook-based routines to trigger from more event sources in the future.

What teams are building

A few common patterns have emerged for early users creating routines:

Scheduled routines

• Backlog management: triage new issues nightly, label, assign, and post a summary to Slack
• Docs drift: scan merged PRs weekly, flag docs that reference changed APIs, and open update PRs

API routines

• Deploy verification: your CD pipeline posts after each deploy, Claude runs smoke checks against the new build, scans error logs for regressions, and posts a go/no-go to the release channel
• Alert triage: point Datadog at the routine's endpoint, Claude pulls the trace, correlates it with recent deployments, and has a draft fix waiting before on-call opens the page
• Feedback resolution: a docs feedback widget or internal dashboard posts the report, Claude opens a session against the repo with the issue in context, and drafts the change

GitHub routines

• Library port: every PR merged to a Python SDK triggers a routine that ports the change to the parallel Go SDK, and opens a matching PR
• Bespoke code review: on PR opened, run your team's own checklist across security and performance, leaving inline comments before a human reviewer looks

Getting started

Routines are available today for Claude Code users on Pro, Max, Team, and Enterprise plans with Claude Code on the web enabled. Head to claude.ai/code to create your first routine, or type /schedule in the CLI.

Routines draw down subscription usage limits in the same way as interactive sessions. In addition, routines have daily limits: Pro users can run up to 5 routines per day, Max users can run up to 15 routines per day, and Team and Enterprise users can run up to 25 routines per day. You can run extra routines beyond these limits with extra usage. See the docs for more information.

Original source

Pair Opus as an advisor with Sonnet or Haiku as an executor, and get near Opus-level intelligence in your agents at a fraction of the cost.

Developers who want to better balance intelligence and cost have converged on what we call the advisor strategy: pair Opus as an advisor with Sonnet or Haiku as an executor. This brings near Opus-level intelligence to your agents while keeping costs near Sonnet levels.

Today we're introducing the advisor tool on the Claude Platform to make the advisor strategy a one-line change in your API call.

Build cost-effective agents with the advisor strategy

With the advisor strategy, Sonnet or Haiku runs the task end-to-end as the executor, calling tools, reading results, and iterating toward a solution. When the executor hits a decision it can't reasonably solve, it consults Opus for guidance as the advisor. Opus accesses the shared context and returns a plan, a correction, or a stop signal, and the executor resumes. The advisor never calls tools or produces user-facing output, and only provides guidance to the executor.

This inverts a common sub-agent pattern, where a larger orchestrator model decomposes work and delegates to smaller worker models. In the advisor strategy, a smaller, more cost-effective model drives and escalates without decomposition, a worker pool, or orchestration logic. Frontier-level reasoning applies only when the executor needs it, and the rest of the run stays at executor-level cost.

In our evaluations, Sonnet with Opus as an advisor showed a 2.7 percentage point increase on SWE-bench Multilingual1 over Sonnet alone, while reducing cost per agentic task by 11.9%.

The advisor tool

We’re bringing the advisor strategy to our API with the advisor tool, a server-side tool which Sonnet and Haiku know to invoke when they need guidance or help with a specific task.

In our evaluations, Sonnet with an Opus advisor improved scores across BrowseComp2 and Terminal-Bench 2.03 benchmarks while costing less per task than Sonnet alone.

The advisor strategy also works with Haiku as the executor. On BrowseComp, Haiku with an Opus advisor scored 41.2%, more than double its solo score of 19.7%. Haiku with an Opus advisor trails Sonnet solo by 29% in score but costs 85% less per task. The advisor adds cost relative to Haiku alone, but the combined price is still a fraction of what Sonnet costs, making it a strong option for high-volume tasks that require a balance of intelligence and cost.

Declare advisor_20260301 in your Messages API request, and the model handoff happens inside a single /v1/messages request—no extra round-trips or context management. The executor model decides when to invoke it. When it does, we route the curated context to the advisor model, return the plan, and the executor continues all within the same request.

response = client.messages.create(
  model="claude-sonnet-4-6",  # executor
  tools=[
    {
      "type": "advisor_20260301",
      "name": "advisor",
      "model": "claude-opus-4-6",
      "max_uses": 3,
    },
    # ... your other tools
  ],
  messages=[...]
)

Advisor tokens reported separately

in the usage block.

Pricing. Advisor tokens are billed at the advisor model's rates; executor tokens are billed at the executor model's rates. Since the advisor only generates a short plan (typically 400-700 text tokens) while the executor handles the full output at its lower rate, the overall cost stays well below running the advisor model end-to-end.

Built-in cost controls.
Set max_uses to cap advisor calls per request. Advisor tokens are reported separately in the usage block so you can track spend per tier.

Works alongside your existing tools.
The advisor tool is just another entry in your Messages API request. Your agent can search the web, execute code, and consult Opus in the same loop.

Get started

The advisor tool is available now in beta natively on the Claude Platform. To get started:

1. Add the beta feature header: anthropic-beta: advisor-tool-2026-03-01
2. Add the advisor_20260301 to your Messages API request
3. Modify your system prompt based on your use case

We recommend running your existing eval suite against Sonnet solo, Sonnet executor with Opus advisor, and Opus solo. Explore the docs to learn more.

Footnotes

1. SWE-bench Multilingual: Sonnet 4.6 solo used adaptive thinking. Sonnet 4.6 + Advisor used our suggested system prompt for coding with thinking turned off. Both runs used high effort with bash and file editing tools. Scores are averaged over five trials of 300 problems across nine languages. Opus 4.6 was used as the advisor model in all runs.

2. BrowseComp: All runs used thinking turned off with web search and web fetch tools. Sonnet 4.6 runs used medium effort. Sonnet 4.6 + Advisor used our suggested system prompt for coding; Haiku 4.5 + Advisor did not. No programmatic tool calling or context compaction. Scores are based on 1,266 problems with one attempt per problem. Opus 4.6 was used as the advisor model in all runs.

3. Terminal-Bench 2.0: All runs used thinking turned off with bash and file editing tools. Sonnet 4.6 runs used medium effort. Neither advisor run used our suggested system prompt for coding. Each task ran in an isolated pod with 3x resource allocation and a 1x timeout. Scores are averaged over five attempts per task across 89 tasks. Opus 4.6 was used as the advisor model in all runs.

Original source

Claude Cowork is now generally available on all paid plans

Within companies, Claude Cowork has become a key part of how teams operate: handling tasks, drafting project deliverables, and keeping teams up to date.

Today, we’re introducing organization controls to help teams deploy Claude Cowork company-wide: role-based access controls for Enterprise, group spend limits, expanded OpenTelemetry observability, and usage analytics for admins to see Claude Cowork adoption.

Early signals

Claude Code helped developers transition from handing Claude questions to whole tasks, and we’re seeing the same pattern across the entire organization with Claude Cowork: the vast majority of Claude Cowork usage comes from outside engineering teams. Importantly, functions like operations, marketing, finance, and legal are not handing Claude their core work, but rather the work that surrounds their most critical tasks—project updates, collaboration decks, research sprints, etc.

As early enterprise adopters of Claude Cowork have seen this pattern emerge in one team, they’ve often wanted to roll it out more broadly, opening questions like who gets access, spend management, and how to see what’s happening across teams.

Controls for organization-wide deployment

Deploying agents with Claude Cowork’s capabilities across an organization requires governance and visibility for admin teams. Today, we’re adding more of the controls organizations need:

• Role-based access controls. Admins on Claude Enterprise can now organize users into groups — manually or via SCIM from your identity provider — and assign each a custom role defining which Claude capabilities its members can use. Turn Claude Cowork on for specific teams and adjust as adoption grows.
• Group spend limits. Set per-team budgets from the admin console. Predictable costs, adjustable as you learn what each team needs.
• Usage analytics. Claude Cowork activity now appears in the admin dashboard and the Analytics API. From the dashboard, admins can track Claude Cowork sessions and active users across various date ranges. The Analytics API goes deeper: per-user Claude Cowork activity, skill and connector invocations, and DAU/WAU/MAU alongside existing Chat and Claude Code figures. See which teams are adopting, which workflows are landing, and where to invest next.
• Expanded OpenTelemetry support. Claude Cowork now emits events for tool and connector calls, files read or modified, skills used, and whether each AI-initiated action was approved manually or automatically. Events are compatible with standard SIEM pipelines like Splunk and Cribl, and a shared user account identifier lets you correlate OTEL events with Compliance API records. OpenTelemetry is available on Team and Enterprise plans.
• Zoom MCP connector. Claude Cowork integrates with the tools your teams already use. Today, Zoom is launching a connector that brings meeting intelligence directly into the Cowork experience. The Zoom connector delivers AI Companion meeting summaries and action items alongside transcripts and smart recordings — helping teams use their conversations on Zoom to create agentic workflows in Cowork. Add Zoom from the connector directory in Claude's settings.
• Per tool connector controls. Admins can now restrict which actions are available within each MCP connector across the organization — allowing read access but disabling write operations, for example. Permissions apply org-wide and are configured from the admin console.

How organizations use Claude Cowork

Zapier connected Cowork to their org database, Slack, and Jira to surface engineering bottlenecks—getting back a dashboard, team-by-team analyses, and a prioritized roadmap that Product and Design Ops then copied for themselves.

Jamf turned a seven-facet performance review into a 45-minute guided self-evaluation, then built similar workflows for vendor reviews and incident response.

Airtree, a venture firm, built a board prep workflow that pulls from a portfolio company's Drive, Slack updates, and competitor news, cross-referenced against the previous prep.

Getting started

Claude Cowork and Claude Code on Desktop are generally available today on all paid plans on macOS and Windows. Download the Claude desktop app at claude.com/download.

For admins deploying Claude across your organization: configure role-based access controls, group spend limits and OpenTelemetry from the admin console. Claude Cowork usage data is available in the admin dashboard, and the Analytics API is documented here.

For a deployment walkthrough, join our April 16th webinar with PayPal.

Original source