Anthropic acaba de romper Claude
Anthropic released three major updates in one week: Opus 4.7 (an improved model), Claude Design (a visual interface for creating web pages, prototypes, and presentations), and Routines (a cloud-based automation system). Together, these updates push Claude toward becoming a complete work environment that can think, design, build, and execute tasks autonomously.
Summary
The video covers three significant Anthropic releases that happened within a single week, framing them as a collective shift in what Claude can do rather than incremental improvements.
The first release is Opus 4.7, an upgrade to Anthropic's most powerful model. The presenter is deliberately skeptical of benchmarks, arguing they don't always reflect real-world performance. Community reactions are mixed — some users notice improvements, others prefer the previous version — and the presenter concludes it's a step forward but not the most revolutionary announcement of the week.
The second and more impactful release is Claude Design, a dedicated visual interface for building web pages, interfaces, prototypes, mobile screens, and presentations. The presenter explains that visual design was one of Claude's clearest weaknesses compared to tools like Google's Stitch. Claude Design addresses this by offering a high-fidelity design mode where Claude asks clarifying questions before generating (similar to Claude Code's planning behavior), produces interactive and editable prototypes, allows direct element-level editing, commenting, drawing annotations, and exporting to formats like PDF, PowerPoint, Canva, or Claude Code. Users can also provide brand assets, fonts, logos, and GitHub links so Claude can maintain a consistent visual identity. The presenter emphasizes this is closer to a Google AI Studio-style prototyping environment than a simple Canva alternative.
The third release is Routines, which the presenter considers the most strategically significant. Routines allow Claude Code automations to run on Anthropic's infrastructure — on a schedule, via webhooks, or API triggers — without requiring a local machine, Mac Mini, or separate server. The presenter demonstrates two examples: a daily email summary routine that reads Gmail, classifies emails by priority, and pre-writes draft replies; and an AI newsletter routine that searches for recent tech and AI news, creates a structured Notion page, and prepares a draft email with a link to that page. The presenter positions Routines as solving the 'middle layer' problem of automation — the orchestration logic that tools like N8N require users to build manually — by letting users describe that logic in natural language instead.
The video concludes by noting that across just one week, Anthropic has addressed model quality, visual design, and autonomous execution — pushing Claude toward covering the full workflow of thinking, designing, building, and running tasks independently.
Key Insights
- The presenter argues that benchmarks for Opus 4.7 are unreliable indicators of real-world performance, noting that models can be optimized to score well on tests without that translating to daily-use improvements — and that community reactions to Opus 4.7 are genuinely split depending on task type.
- The presenter explains that Claude Design's key differentiator over a normal chat prompt is not the underlying code generation but the visual layer — being able to see options, compare results, and interact with a design interface rather than going 'blindly with just context.'
- Claude Design's questioning behavior before generating — asking about visual metaphors, metrics, interactions, and controls — is compared to Claude Code's planning-before-building approach, which the presenter argues is critical for preventing generic outputs.
- The presenter frames Routines as solving the 'middle layer' of automation: where tools like N8N require users to manually build nodes, credentials, authentications, and variables, Routines lets users describe that orchestration logic in natural language and have Claude execute it in the cloud.
- The presenter positions the three releases together as evidence that Anthropic is building Claude into a complete work environment covering the full flow — thinking, designing, building, and executing — rather than just improving a chat model.
Topics
Transcript
[0:00] Antropic has been going on for a crazy week because they haven't released the typical small new feature to say that Cloud is improving a little and that's it. They've released several things that really expand what you can do with the tool. On one hand, they have launched Opus 4.7, a new version of their most powerful model. On the other hand, they have launched Cloud Design, so now Cloud can get much better involved in all the visual aspects, designing interfaces, creating screens, prototypes, and giving much more [0:30] aesthetic sense to what it creates. And they've also launched routines, which basically means Cloud is starting to get very seriously involved in the field of automation. And…
Full transcript available for MurmurCast members
Sign Up to AccessMore from Migue Baena IA
Cómo aprender por tu cuenta y volverte imparable con IA
This tutorial demonstrates how to use Gemini Notebook (formerly Notebook LM) to self-teach any skill by curating personal learning sources and converting them into multiple study formats. The four-step process involves selecting quality materials, studying them through different media, teaching the content to others, and integrating learning into daily routines.
Tienes que probar esto antes de crear tus videos con IA 🎬
A new AI video creation tool called 3D Builder within Topw Film Studio allows creators to pre-visualize and plan movie scenes by building interactive 3D environments, positioning characters and objects, controlling camera movements, and reviewing animations before final video generation.
No elijas entre Claude y Codex: úsalos juntos 🤝🏻
The video proposes using Claude and Codex together rather than choosing one over the other to maximize their respective strengths. Claude handles problem analysis and task delegation while Codex executes code generation, allowing users to work around individual API rate limits and leverage complementary capabilities.
Cómo usar Claude mejor que el 99% de la gente
A comprehensive guide to using Claude AI effectively, covering account setup with context import, model selection and reasoning levels, proper prompting techniques, projects and skills for organizing work, and advanced features like Cowork for delegated tasks and Code for building applications.
El archivo de GitHub que deberías de conocer 👀
Anthropic has published a detailed GitHub repository containing a 275,000+ character PRD with nearly 8,000 lines of instructions for one of their models, including specifications on what to do, avoid, and how to use tools. The key value lies not in copying the content but in studying its structure and learning how it organizes rules and defines tool usage for prompt engineering applications.