Three developments this week show how fast software moves — and where the gaps still exist. A JavaScript runtime rewrites itself for performance, users get early access to ChatGPT’s next version, and research reveals a critical flaw in AI document handling.

Bun’s Rust Rewrite Hits 99.8% Compatibility

Bun’s experimental Rust rewrite reached 99.8% test compatibility on Linux x64 glibc systems. The JavaScript runtime, originally written in Zig, is being rebuilt in Rust for better performance and memory safety.

This matters because runtime performance directly affects your application costs. Faster JavaScript execution means fewer server resources, lower cloud bills, and better user experience. The 99.8% compatibility rate suggests the rewrite maintains existing functionality while improving the foundation.

For businesses running Node.js applications, this could mean significant infrastructure savings. When we build custom solutions at Artemis Lab, runtime efficiency translates to real cost differences — especially at scale.

ChatGPT 5.5 Pro Goes Live

Users report accessing ChatGPT 5.5 Pro, apparently the next iteration beyond GPT-4. Early feedback suggests improved reasoning and code generation capabilities, though OpenAI hasn’t officially announced the release.

The pattern here is incremental AI improvements hitting production faster. Each model generation handles more complex tasks with better accuracy. This means AI agents can take on work that previously required human oversight.

Businesses using AI for customer service, content generation, or data analysis should expect capabilities to keep improving rapidly. What required custom development six months ago might be available through standard API calls today.

LLMs Corrupt Documents When Delegated

New research shows LLMs introduce subtle errors when given document editing tasks. The corruption isn’t obvious — it’s small changes that alter meaning while maintaining readability.

This is a critical issue for businesses using AI for document processing. Legal contracts, technical specifications, and financial reports can’t tolerate “helpful” AI modifications. The research confirms what many suspected: AI shouldn’t have unsupervised access to important documents.

Smart implementation means using AI for drafting and analysis, not final document handling. When we build RAG systems for clients, we design workflows where AI suggests changes but humans approve them. The research validates this approach.

The broader lesson: AI capabilities are advancing fast, but so is our understanding of their limitations. The companies that succeed will be those that use AI strategically, not universally.