A developer has demonstrated the creative power of AI by using Anthropic's Claude to build a complete programming language from scratch. The language, called "Cursed," is uniquely based entirely on Gen-Z slang.

The project was completed by someone with no prior compiler experience, highlighting how AI enables individuals to tackle previously out-of-reach complex tasks. What does it mean for software development when anyone can generate not just code, but entire programming ecosystems?

Today in AI Brief:

  • Claude creates a Gen-Z programming language

  • Mistral’s chip giant partnership

  • Claude now creates spreadsheets and decks

  • The security risks of on-device AI agents

AI creates a programming language

In Brief: In a wild display of AI-assisted creation, a developer put Anthropic's Claude in a continuous loop for three months to generate a fully functional programming language called Cursed.

The Details:

  • The language was created by a developer without prior compiler experience, who used a simple prompt to guide Claude through the entire process from concept to a working compiler.

  • The result is not a toy but a complete language with interpreted and compiled modes, editor extensions for VSCode and Vim, and an open-source GitHub repository.

  • True to its theme, the language's entire lexical structure is based on Gen-Z slang, with keywords like slay for functions, facts for constants, and vibe_check for a switch statement.

Take Away:

This experiment shows how modern AI can act as a powerful co-creator, enabling a single person to build complex systems that were previously out of reach. It also offers a glimpse into a future where developers use AI to generate not just code, but entire software ecosystems and tools.

Mistral AI's chip giant partnership

In Brief: French AI leader Mistral AI announced a €1.7B Series C funding round and a strategic partnership with ASML, the world's leading semiconductor equipment manufacturer.

The Details:

  • ASML led the funding round with a massive €1.3 billion investment, securing an approximate 11 percent stake in the French AI company.

  • The collaboration will integrate Mistral’s AI into ASML’s holistic lithography products to help ASML's customers achieve faster time-to-market and higher performance.

  • Beyond a simple vendor relationship, ASML's CFO will join Mistral AI's Strategic Committee, giving the chip leader an advisory role in Mistral's future direction.

Take Away:

This partnership signals a major move to embed frontier AI directly into the core of the semiconductor value chain. It gives Mistral AI deep industrial access while providing ASML a crucial edge in optimizing the complex chip manufacturing process.

Claude Now Creates Spreadsheets and Decks

In Brief: Anthropic just gave Claude a major upgrade, allowing it to create and edit files like Excel spreadsheets, PowerPoint decks, and PDFs directly from your prompts.

The Details:

  • Claude accomplishes this by using a private computer environment to write and execute code, effectively turning text prompts into ready-to-use documents and analyses.

  • You can now transform raw data into polished charts, build complex financial models, or even convert a PDF report into a PowerPoint slide deck just by asking.

  • This feature is rolling out as an experimental preview for Max, Team, and Enterprise users, and you must enable it in settings to get started.

Take Away:

This update transforms Claude from a conversational advisor into an active digital collaborator that executes complex tasks. It significantly closes the gap between an idea and a tangible output, making powerful analysis accessible without needing to code.

The Risk of AI Agents on Your Laptop

In Brief: A new analysis argues that giving AI agents full control over your laptop is a significant security risk because modern operating systems are not built to safely contain them.

The Details:

  • The core issue is that desktop operating systems lack the granular controls to block an AI agent’s access to sensitive data—like password managers or credential files—exposing users to risks from the lethal trifecta of prompt injection.

  • One promising solution is moving agents to sandboxed cloud environments, which offer better security and reproducibility, a direction that Claude Code is working on.

  • The other potential fix is browser integration, using the browser’s built-in sandboxing to isolate agent actions, but this has proven difficult to implement safely.

Take Away:

The convenience of on-device AI agents comes with security trade-offs that today’s operating systems can’t fully mitigate. The future of secure and powerful agent workflows will likely live in the cloud or be tightly woven into the browser itself, not run with free rein on your local machine.

Everything else in AI

Anthropic advocated for thoughtful AI regulation while officially endorsing California’s SB 53 bill, which mandates safety frameworks and public transparency reports from frontier model developers.

The Frontier Model Forum announced a first-of-its-kind information sharing agreement to help top AI labs collaborate on safety research and risk mitigation for powerful models.

Google DeepMind introduced its Frontier Safety Framework, a set of procedures for evaluating and mitigating risks in advanced AI that aligns with the principles in California's proposed SB 53 bill.

That’s all for today!