What's new at Anthropic: March roundup with my take

Anthropic has been going full throttle lately. Here's a roundup of what happened in March 2026 — with my take on each piece of news.

Claude 4 — a new generation of models

The biggest news of the month. Opus 4 is the most powerful model I've ever tested — reasoning, code, long context, everything at a new level. Sonnet 4 is the sweet spot for daily work. I wrote a detailed review in a separate article.

My take: After a week of testing, I switched everything to Sonnet 4. I only pull Opus 4 out for architectural tasks. Sonnet's price is reasonable and the quality is better than Opus 3. An unambiguous upgrade.

Claude Code gets hooks and superpowers

Probably the biggest news for me personally. Claude Code now supports hooks — automated actions that fire on certain events. For example:

• Automatically run linting before every commit
• Check types after editing a file
• Load project context from CLAUDE.md on session start

My take: This is a game changer for consistency. I used to occasionally commit something that wouldn't pass lint. Now the hook catches it automatically. And the CLAUDE.md file — it's like a README for AI. You describe the project in it and Claude Code reads it on every session. Genius in its simplicity.

On top of that, skills have been added — prepared instructions for specific tasks. Brainstorming, code review, test-driven development. Claude Code is going from a terminal tool to a fully-fledged developer environment.

MCP: OAuth and remote servers

MCP got OAuth authentication for remote servers. That means you can securely connect Claude to cloud services — not just local servers.

My take: This is what I was waiting for. Local MCP servers were great for development, but for production deployments you needed secure authentication. OAuth solves 90% of enterprise use cases. The remaining question is the audit log — who read what and when? Anthropic is reportedly working on it.

Vercel AI Gateway with Claude

Vercel launched AI Gateway with native Claude support. One API, automatic failover between providers, cost tracking, zero data retention.

My take: This is exactly what I needed. Instead of calling the Anthropic API directly, everything now goes through the gateway. Benefits:

• Automatic fallback to another model when the Claude API has an outage
• Cost overview in one place
• OIDC authentication — no API keys in .env files

For side projects like this site it might be overkill. For production applications it's a must-have.

Claude for enterprise: SOC 2 Type II

Anthropic earned SOC 2 Type II certification. For nerds: that means an independent auditor verified that Anthropic has working security controls — not just on paper, but in practice for at least 6 months.

My take: This is boring news that's actually huge. Lots of companies couldn't deploy Claude because the compliance team said "they don't have SOC 2." Now they do. I'm expecting a wave of enterprise adoptions in Q2.

Artifacts 2.0

Claude.ai got improved Artifacts — interactive outputs you can share as a link. Including:

• React components (live preview)
• Mermaid diagrams
• Interactive tables and charts
• Markdown documents with PDF export

My take: I use Artifacts for quick prototypes. "Make me a React component for a pricing table" → click → live preview → share the link with the client. It's like CodePen, but AI-powered. Version 2.0 is faster and more stable.

What I'm expecting in April

Based on what Anthropic has been hinting at:

1. Claude Code in IDE — extensions for VS Code and JetBrains, not just the terminal
2. Improved vision — better analysis of screenshots and designs
3. Longer context for Sonnet — Sonnet 4 could get 500k+ context

The last one would be huge — Opus with 200k is great, but Sonnet with half a million tokens would cover nearly every use case at a third of the price.

Summary

In March 2026, Anthropic showed they're not sleeping:

| News | Impact for developers | |---------|-------------------| | Claude 4 (Opus/Sonnet/Haiku) | Highest — generational leap | | Claude Code hooks | High — workflow automation | | MCP OAuth | High — enterprise integration | | AI Gateway (Vercel) | Medium — easier deployment | | SOC 2 Type II | Medium — enterprise adoption | | Artifacts 2.0 | Low — nice to have |

Overall? A good month for the Claude ecosystem. And I enjoy watching Anthropic gradually build not just a model, but a whole platform around it. That's what makes them more than just an "AI company" — it makes them a developer platform.

See you at the April roundup.