Claude Commerce Agents is now open-source: a blueprint for building shopping and merchant agents with reference implementations across retail, travel, telecom, and entertainment.
Retailers using Claude-powered shopping agents have seen average cart sizes grow 35% and purchase completion rates rise 60%. The blog covers architecture, latency and cost techniques, and eval practices behind those results.
Fable 5.1 is recommended for lower-effort settings on tasks with fewer edge cases or less need for verification. A notable fix: switching effort levels no longer breaks prompt cache, which was a previous limitation.
A user on a 20x Pro plan hit their 5-hour session limit after roughly 45 minutes of coding across three chats, also consuming 38% of their weekly Fable allocation. A team member asked them to run /feedback in the affected sessions and share the feedback ID.
Computer use in the Claude Code desktop app now runs in the background on macOS. Claude can operate permitted apps while you keep working. It's enabled automatically for existing computer use users, or can be turned on via Settings > General. Currently in beta for Pro and Max subscribers.
Fable 5.1 is now available in Claude Code, Claude Cowork, and Claude Tag. The team framed the launch around taking on ambitious, previously months-long projects and is soliciting community feedback.
Fable 5.1 was observed proactively adding ruff lint rules to prevent regressions mid-task. A team member confirmed this is an intentional, valued behavior: the Claude and Claude Code codebases use many custom lint rules as guardrails, and models have been good at writing them since the Sonnet 3.6 era.
Users with DISABLE_TELEMETRY or CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC set in their settings.json won't receive Fable 5.1 configs. The call that delivers those configs is the same one the env var blocks, by design, because it could theoretically be used for telemetry.
Long-running Fable 5.1 sessions are supported indefinitely. bcherny confirmed personally running it for weeks.
A Japanese user noted that Claude Code version 2.1.257 shipped with 104 changelog entries. A team member replied asking which change was their favorite.
A user running three high-effort and two medium-effort Fable 5.1 threads burned through 12% of their weekly limit and 40% of their 5-hour limit in 18 to 20 minutes. A team member reached out via DM.
A user reported their limits disappeared in a single prompt, flagging it as likely a bug. A team member asked them to run /feedback in the affected session and share the resulting ID.
A user reported their 5-hour usage limit was exhausted in under 30 minutes with Fable 5.1. A team member asked them to run /feedback when it happens again.
Artifacts can now call MCP connectors directly. Share a dashboard and every viewer's copy pulls fresh data through their own connections, with no refresh jobs, no shared credentials, and nothing to maintain.
Documentation is now available for the new Artifacts feature that lets shared dashboards pull live data through each viewer's own MCP connectors.
