Tag
All content about Claude, organized for fast scanning.
Claude has introduced new versions, Haiku 5.5 and Sonnet 5.5, which focus on enhancing performance and reducing costs for high-volume tasks and everyday applications, respectively. Additionally, Claude Code's effort settings allow users to adjust the balance between speed and verification depth in code-related tasks, while a forthcoming plugin portal aims to streamline submissions and analytics for developers. Overall, these developments reflect a commitment to improving efficiency and user experience across the Claude platform.
ClaudeDevs has just rolled out Claude Haiku 5.5, targeting high-volume workloads like summaries, compactions, and database queries at claimed lower average cost. Max and Team subscribers also get monthly Claude Platform API credits, up to $500 pooled for Team.
Anthropic has just rolled out Claude Sonnet 5.5, bringing a faster Sonnet model built for everyday tasks like bug fixes, docs, slides, and UI work. The company says it’s 30% faster and can cost up to 30% less per task while improving key benchmark scores.
Thariq’s tests suggest Claude Code’s effort slider mostly changes verification, edge-case testing, and independent judgment—not just response length. Low/medium shine for fast iteration, while high/max can pay off for security, code review, and tricky benchmarks.
Thariq breaks down how Claude Code’s effort levels change behavior, from quick, collaborative edits to higher-effort runs that challenge assumptions and test edge cases. His benchmarks show effort boosts verification, but won’t reliably fix a wrong approach.
Claude is preparing a plugin portal for submissions, review tracking, and usage insights, based on a post shared by Boris Cherny quoting ClaudeDevs. The group says plugins bundle MCP and skills and claims MCP usage is up 110x this year, without data.
Thariq now plans to preserve Claude’s plan mode as a customizable built-in mod after user feedback. The proposal would let you tweak the plan prompt, add or share new modes, and even override or rebind the Shift+Tab shortcut.
Claude has just rolled out Opus 5.5, bringing claimed gains in agentic coding, computer use, and knowledge work—while cutting run costs about 40% versus Opus 5. The model is also said to be 30% faster, with updated writing behavior and reduced token pricing.
Claude Code now supports AGENTS.md as a fallback when CLAUDE.md isn’t present, starting in version 2.1.277. The behavior can be toggled via /config and is powered by upcoming Claude Code mods for deeper customization.
Anthropic has just rolled out Claude Fable 5.1 and Claude Mythos 5.1, touting stronger coding and knowledge-work performance. The company also says cache reads and Claude Code sessions are cheaper, with fewer bio- and cyber-safety interruptions.
Claude has just rolled out Fable 5.1 and Mythos 5.1, positioning them for coding, research, cybersecurity, and life sciences. The company touts higher Terminal-Bench and OSWorld scores versus prior models, plus lower cache-read costs and new Enterprise Frontier Safeguards.
Anthropic will adjust Claude Code’s weekly usage limits on September 14, replacing today’s 50% promo bonus with a permanent 25% increase for paid plans. The company says it’s about sustainable capacity, but users are calling it a cut.
Anthropic has just rolled out Claude Mythos 5 in Claude Security, now in public beta for all Claude Enterprise customers. The service scans GitHub repos for vulnerabilities and suggests fixes, but Mythos 5 isn’t exposed for direct use. Pricing is standard token usage.
Claude has just rolled out a reliability upgrade for Claude Code Remote Control, including the ability to start new sessions from your phone. It also aims to keep mobile and desktop sessions in sync and fix stale “offline” status. Some users still report dropouts and Android quirks.
Claude Code has just rolled out a new Concise output style, aimed at leading with results and trimming overly long responses. It’s available via /config and settings.json, and comes after growing user complaints about verbosity and context bloat.
Claude Code has just rolled out a new /design research preview, bringing Claude Design-style artboards to the CLI and desktop app. Subscribers can generate editable UI concepts, pick an option, and ask Claude to implement it—though access, overlap, and token cost remain open questions.
Claude Code is rolling out auto mode as the default permission setting for Pro, Max, and Team users. It uses a separate safety classifier to review commands instead of prompting each time, with confirmations for existing custom defaults.
Boris Cherny is testing a Claude maintenance workflow in Slack that runs daily across iOS, Android, desktop, web, CLI, and SDK repos. The routines fuzz for crashes, fix flaky tests, prune dead code, and more—opening 388 PRs, with 180 merged so far.
Claude Code has just added cross-session messaging, letting one task ask another questions and send updates. Instead of syncing full chat logs or files, it hands off a generated summary to reduce repeated setup. The feature is available on macOS and Linux.
Claude Code has updated /fork to spin up a background session while you keep working. Its former subagent behavior now lives in /subtask, which reports a final result back. The changes are live in v2.1.212+.
Anthropic’s Thariq shared new context-engineering guidance for Claude 5, arguing prompts are only part of what the model sees. The company says it cut 80%+ of Claude Code’s system prompt with no measurable loss, leaning more on tools, Skills, and references.
Anthropic has just rolled out Claude Opus 5, touting near-frontier performance at half the price of Fable 5. The company’s benchmark bundle puts it near the top in coding, knowledge work, and agentic tasks—though results vary by test.
Boris Cherny lays out a four-step blueprint for AI adoption, focused on breaking bottlenecks and layering guardrails—not just adding tokens. He highlights features like automated reviews, multi-agent workflows, and self-verification, plus a better ROI test than usage metrics.
Boris Cherny argues teams should make agentic coding useful by encoding repeat work into lint rules, CI, tests, and repo docs—not one-off fixes. He suggests files like CLAUDE.md and REVIEW.md so agents can contribute with minimal prompting.
ClaudeDevs has just rolled out an in-app browser for Claude Code on desktop, so it can read and interact with docs, designs, and websites without switching apps. The browser is sandboxed and configurable, with optional session persistence.
Claude Code is getting a new /checkup maintenance command aimed at tidying unused skills, MCPs, and plugins. It can also deduplicate and split CLAUDE.md files, disable slow hooks, and update settings to reduce per-session context overhead.
Boris Cherny, creator of Claude Code, discusses his path from startups and Instagram to Anthropic, where Claude Code emerged from the company’s focus on coding, enterprise, and safety. He explains how AI agents are transforming software development, why businesses must reorganize around AI to unlock major productivity gains, and how Anthropic is expanding compute and developer tools. He also reflects on the future of jobs, software engineers, startups, security, and why he believes coding will become a basic skill like reading and writing.
The opening of the talk defines “vibe coding” as more than just using AI to help write code. The speaker argues that true vibe coding means letting the model handle the implementation to the point that you “forget the code exists,” while you focus on the outcome. He explains why this matters: as AI systems get better, they will be able to handle larger and larger chunks of work, making it unrealistic for humans to stay in a tight line-by-line review loop forever. He then frames the core challenge as how to use this approach safely in production. His answer is that engineers should stop obsessing over every implementation detail, but still stay accountable for the product’s behavior and quality. He compares this to managers or executives overseeing work they cannot personally execute in full detail: they succeed by verifying outcomes, requirements, and checkpoints rather than inspecting everything directly. A key caveat in this early section is tech debt. He says that unlike product behavior, tech debt is still hard to validate without actually understanding the code. Because of that, he recommends using vibe coding mainly on leaf nodes of a codebase, meaning isolated features where problems are less likely to spread into the core architecture.
Agents can be smart and still unreliable for real work when they lack domain expertise. This talk argues that the next step isn’t more agent scaffolding—it’s packaging reusable expertise as “Skills” that agents can load when needed. Key takeaways Breaks down what Skills are: organized folders of procedural knowledge (including scripts/tools) that can be versioned, shared, and composed. Explains why code and the file system can act as a universal interface, while Skills supply the missing domain context. Walks through “progressive disclosure” to protect the context window and enable libraries of hundreds or thousands of Skills. Maps an emerging architecture: agent loop + runtime environment + MCP servers for connectivity, with Skills providing expertise—and how that supports deployment across domains.
Coding agents are everywhere—but what actually made them finally usable for real engineering work? This talk walks through Jared Zoneraich’s independent breakdown of Claude Code’s architecture and why he believes “simple” beats brittle, over-engineered agent frameworks. Key takeaways Breaks down the “Master Loop” approach: a single-threaded while-loop that repeatedly calls tools and feeds results back to the model. Covers the core toolset (e.g., Bash, Grep/Glob, diff-based editing) and why Bash is central for robustness and training-data reasons. Explains prompt-driven planning via to-do lists—structured guidance without deterministic enforcement. Discusses reliability concerns: sandboxing/permissions, context management, and the use of sub-agents to avoid bloating the main context.
If you’re weighing Claude Code versus Cursor (or trying to juggle both), this walkthrough shows what changes when Claude Code is fully integrated inside Cursor—so you can work with live context without constantly re-explaining your code. Key takeaways Walks through setup in Cursor’s integrated terminal: updating Claude, signing in, and confirming the integration. Shows how the status indicator reflects the active file, enabling prompts like “what does this file do?” without naming it. Demonstrates line awareness: select a block of code and ask “what does this do?” with the selection as context. Breaks down the diff view workflow—review proposed changes, revert specific blocks, and accept/reject (with an optional auto-accept toggle), plus linking Cursor from an external terminal via /ide.
Learn how to integrate Claude Code 4 with Cursor in this step-by-step tutorial. This new feature from Anthropic brings full Claude Code support to the Cursor AI IDE, giving you new features, like inline diff views and real-time code context, all within your favorite IDE. I’ll walk you through the setup process for Cursor and explain key features that make this integration of the best AI coding agent upgrades yet.