Claude Code adds auto mode for parallel sessions

• According to ClaudeDevs' rollout post, Claude Code's auto mode now reaches the Pro plan and supports Sonnet 4.6 alongside Opus 4.7, expanding a mode that removes permission prompts and lets the agent keep running.
• bcherny's workflow note turned that feature into the real headline: auto mode is what makes "multi-clauding" viable, with one session running while you open another in parallel.
• The creative demos arrived fast. kvnkld's Empty State and Navigation clip, viktoroddy's Claude Code plus Seedence 2.0 thread, and a run of markproduct website demos all showed UI sections, landing pages, and animated sites assembled in minutes.
• The rough edges showed up just as quickly. In Everlier's profiling test, Claude Code burned time and tokens without finishing a startup benchmark task, while levelsio's screenshot showed effort settings getting pushed back to medium.

Anthropic linked the rollout to its permission-modes docs, where auto mode is positioned as the prompt-killing path for longer runs. You can also jump from viktoroddy's tutorial into a full animated-site workflow, then compare that polished path with Everlier's failed benchmark run and levelsio's effort-reset complaint, which is a very different picture of what happens once the task stops being a clean website build.

The official change is small but potent: auto mode is now on Pro, and it runs with Sonnet 4.6 as well as Opus 4.7. Anthropic's linked permission-modes documentation frames it around eliminating approval popups, which is the boring-sounding feature that changes how long a session can stay useful.

That matters in practice because the old permission rhythm kept turning Claude Code into a supervised tool. Auto mode shifts it toward longer agentic runs, especially for repetitive build steps, UI iteration, and cleanup work.

Boris Cherny, Claude Code lead at Anthropic, described auto mode as his top tip for getting value from Claude Code, because it lets one session keep running while you work in another. His phrase, "multi-clauding," is the cleanest description of the workflow this update unlocks.

Cherny's earlier post also pushed a second point: Anthropic is trying to make agent use feel more normal, not more elite. That fits the rollout. Moving auto mode down to Pro turns a power-user habit into something a much larger slice of designers and indie builders can actually try.

The most convincing evidence for the update is not a feature list. It is the flood of fast, visible outputs.

• kvnkld's demo showed empty-state and navigation work built with Claude Code.
• viktoroddy's thread paired Claude Code with Seedence 2.0 and Motionsites for animated website production.
• markproduct's landing page clip, another five-minute build, and an entire website demo all leaned on the same pitch: short sessions, polished frontend output, and a workflow that looks closer to directing than hand-coding.

The creative angle here is concrete. Claude Code is showing up less as a pure coding assistant and more as the orchestration layer that glues promptable media tools, site builders, and frontend edits into one fast loop.

The clean demos hide the part practitioners care about most: whether the tool holds up on messy tasks. Everlier asked Claude Code to profile Zed and VS Code startup time side by side with hyperfine, and the session reportedly spent tokens and time without completing the assignment.

A separate complaint from levelsio's post pointed at effort settings being forced back to medium, with the author attributing it to Anthropic's GPU shortage. That does not contradict the glossy website demos. It explains them. Claude Code looks strongest when the task is bounded, visual, and easy to evaluate, and shakier when the job turns into benchmarking, persistence, or resource-heavy runs.