Get Started with Claude Managed Projects & New AI Features

What's New with Claude: A Hands-On Guide to the Latest Features

Anthropic has been busy. In the past few weeks alone, the company behind Claude has pushed out a new model release, rolled out a powerful enterprise feature called Managed Projects, published a deep-dive into how Claude operates across web, code, and collaborative environments — and pledged another $20 million toward AI safety research. If you're a developer, team lead, or just a Claude power user, now is the perfect time to level up your workflow.

What Is Managed Projects?

The biggest news for teams is Claude's new Managed Projects feature — currently in testing and rolling out to enterprise users. Think of it as a dedicated project workspace where Claude maintains persistent context, memory, and custom instructions across an entire project lifecycle. No more re-explaining your codebase or style guidelines every time you start a new chat.

Here's how to set it up:

• Create a project space. In the Claude dashboard, navigate to the Projects tab and click "New Managed Project." Give it a name tied to your actual work — "Frontend Refactor Q3" rather than "Test."
• Upload project context. Drag in your key documents: architecture diagrams, API specs, coding standards, and existing code files. These become part of the project's persistent knowledge base.
• Set custom instructions. Dial in Claude's behavior. Tell it your tech stack, preferred patterns, and conventions. Example: "Always use TypeScript strict mode. Never use any. Prefer functional components."
• Invite collaborators. Managed Projects supports team access. Each member shares the project's knowledge while keeping their own session context.

Once configured, every conversation inside the project inherits this context automatically. Claude's answers are more consistent, require less correction, and stay aligned with your project's conventions from the first message.

Putting Claude's Multi-Context Skills to Work

Anthropic recently detailed how Claude maintains coherence across three distinct domains: web interactions, code generation, and collaborative cowork. Understanding these modes helps you pick the right approach for each task.

Web Mode. Claude processes and reasons about web content — documentation, articles, APIs — and can summarize, compare, and extract structured data. Use this for research-heavy tasks like competitor analysis or digesting long technical proposals.

Code Mode. This is Claude's sweet spot. With Managed Projects, Claude remembers your entire codebase structure, not just the snippet in the current window. It can suggest multi-file refactors, spot inconsistencies across modules, and generate boilerplate that follows your established patterns.

Cowork Mode. Designed for real-time collaboration. Use it for brainstorming architecture decisions, reviewing PR descriptions, or debugging sessions with a colleague.

A $20 Million Safety Bet

In parallel, Anthropic announced a $20 million top-up to its AI safety funding — bringing its total commitment to well over $100 million. For users, this isn't abstract philanthropy. The research trickles down into refusal behavior tuning, constitutional AI guardrails, and prompt injection resistance.

Practically, this means:

• Claude is less likely to hallucinate on complex, high-stakes tasks.
• Refusal boundaries are more granular — fewer false rejections on legitimate coding questions.
• Long-running conversations stay on the rails with less contradictory responses.

Getting Started with the New Claude Model

Earlier this year, Anthropic quietly rolled out a "Mythos-like" model to select Wall Street firms, where it made a significant impression. That model is now available to the public, bringing improvements in reasoning depth, code generation accuracy, and long-context comprehension.

To access it in your workflow:

1. In the Claude interface, select the latest model version from the model picker.
2. If you're using Managed Projects, the new model is automatically available — no migration needed.
3. For API users, update your model identifier to the latest stable version in Anthropic's documentation.
4. Test on a sample of your production prompts. The model shines on multi-step reasoning tasks like SQL generation, complex regex, and multi-file refactoring.
5. Use Managed Projects' custom instructions to steer behavior back if you see regressions.

The model handles ambiguous instructions much better. Give it a loosely scoped request like "clean up the error handling in this module" and it will proactively identify gaps, suggest specific try-catch placements, and flag existing patterns that swallow exceptions silently. That initiative is a real step forward.

Tying It All Together

The through-line is clear: Anthropic is betting that context is the killer feature. Managed Projects gives you persistent context. The model improvements give it deeper reasoning. The safety research ensures context doesn't lead to harmful behavior.

Your action plan: enable Managed Projects if you have access, review your custom instructions (the single highest-leverage thing you can do), and try the new model on a task you previously found difficult. The pace of AI change is relentless, but one takeaway is simple: invest time in setting up context and instructions, and Claude will reward you with better results every time.