Anthropic is steadily refining Claude Code, the AI coding tool that Bloomberg describes as "the hottest piece of software in the world" in an interview with its creator.
The attention reflects a broader shift in how software gets written. Several new tweaks and explainers have surfaced around the product in recent days.
According to TestingCatalog, Anthropic is testing a new placement for Projects on Claude Desktop, an interface change aimed at how users organize their work.
On the tuning side, quasa.io reports that Anthropic has explained when developers should adjust the "effort level" setting in Claude Code, and the guidance is essentially: not often. That advice suggests the tool is meant to work well on its defaults, with manual tinkering reserved for edge cases rather than everyday use.
A recurring theme across the coverage is the idea of a "context-rich" AI coding harness — the software layer that feeds an AI model the surrounding information it needs to write useful code. Ars Technica, in a piece featuring Augment Code's Vinay Perneti, makes the case that going "beyond grep" (the classic text-search tool) toward richer context handling is central to how these systems work, framing the discussion around models, harnesses, and context.
Crypto Briefing extends that argument to a specific domain, reporting that Anthropic's context-rich coding harness could reshape how smart contracts — the self-executing code behind many blockchain applications — get built.
Taken together, the items point to a product being polished on multiple fronts at once: interface, configuration guidance, and the underlying approach to feeding AI the right context.
Why it matters: as AI coding tools like Claude Code move from novelty to daily infrastructure for many developers, small changes to how they are configured and how they gather context can meaningfully affect the quality and reliability of the software being built.