Keep pulling the thread on Boris Cherny.
Anthropic engineers cannot see customer conversation data, even when it would be useful for debugging a bug in Claude Code.
Many banks are using Claude Code for migrating their legacy COBOL codebases.
A developer on the Bun team used Claude Code to migrate the entire Bun codebase from the Zig programming language to Rust in about 11 days.
The Bun codebase migration from Zig to Rust cost approximately $150,000 in compute credits.
Anthropic has solved the memory problem for its models, allowing them to maintain coherence over long periods, with some sessions running for weeks at a time.
The explosion in Claude Code's popularity in 2026 is almost entirely due to improvements in the underlying models, not the user interface or harness.
The recent explosion in Claude Code's adoption is almost entirely due to improvements in the underlying models, not the user interface.
Claude Code's growth saw clear inflection points following the releases of Opus 4 in May of last year, Opus 4.5 in November, and Opus 4.6 in February.
In a security competition offering a $20,000 prize, external researchers were able to prompt inject every participating model except for Anthropic's model within Claude Code.
Anthropic's model resisted prompt injection due to a combination of alignment work, mechanistic interpretability probes that detect attacks in the model's neurons, and the "auto mode" permission system in Claude Code.
Since the release of Opus 4.7 or 4.8, and especially Fable, the code generated by the model is almost always better than the code Boris Cherny would have written himself.
With features like loops, routines, and Claude Tag, software development is moving to a higher level of abstraction where a user talks to a model, which in turn talks to other models that write the source code.