Claude Code now runs coding tasks without your approval

Most AI coding tools still ask permission before they act. On 31 May 2026, Anthropic changed that. Claude Code's new Auto mode lets the assistant complete routine programming tasks in the background, interrupting developers only when something genuinely unusual happens. That shift is smaller than it sounds in a press release and larger than it sounds in a sprint meeting.

What Anthropic actually shipped

Claude Code is Anthropic's command-line developer assistant. Until this update, it operated in a confirm-first model: every non-trivial action required a developer to approve it before the tool proceeded.

Auto mode removes that default gate for routine tasks. The assistant now runs background safety checks continuously. It surfaces an interruption only when it detects an action outside a defined risk threshold. Alongside Auto mode, Anthropic also shipped granular cost-tracking tools and automated code review commands as part of the same release.

Access is through the Claude Code CLI. Pricing and tier requirements for Auto mode were not confirmed in the release notes at time of writing.

Why this matters for UK development teams

For developers at UK SMEs, the practical shift is about cognitive load, not raw speed. Constant approval prompts fragment concentration. A developer approving twenty routine commands per hour is not reviewing code, they are managing a tool.

Auto mode addresses that directly. Background checks handle the low-risk work. The developer's attention is reserved for decisions that actually need a human.

For IT managers, the new cost-tracking tools are the more immediately useful addition. API costs for AI coding tools can accumulate quickly and unpredictably. Granular tracking per command or session gives IT managers the visibility to set meaningful budgets and identify which workflows are consuming resource.

The autonomous capability removes the need for manual approval of routine coding tasks, and the updated system integrates background safety checks with cost-tracking tools that connect to existing monitoring dashboards. That combination reduces administrative overhead across a technical team, not just for individual developers.

AI adoption is 80% people and culture, 20% technology. The biggest risk with Auto mode is not the tool acting incorrectly. It is a team deploying it without agreeing what the risk threshold settings should be, or without a shared understanding of when a human review gate still matters.

Three actions for this week

1. Review the Auto mode risk threshold documentation in the Claude Code release notes before enabling it in any production environment. This takes under 30 minutes and ensures the team understands what triggers an interruption before that decision is delegated to the tool.

2. Ask each developer using Claude Code to run one sprint's worth of tasks with Auto mode enabled in a staging environment. Log the interruption rate. If genuine risk flags are rare, that is evidence the tool is calibrated correctly for your codebase. If interruptions are frequent, the threshold settings need adjusting before moving to production.

3. Configure the new cost-tracking feature before the end of this week. Set a per-project or per-team budget ceiling. Assign an IT manager to review the usage dashboard at the end of each sprint. This step requires no additional licence; it is part of the same update.

Considerations and limitations

Auto mode is only as safe as the risk threshold it is given. A poorly configured threshold either creates the same volume of interruptions as the old model, or misses actions it should flag. Neither outcome is acceptable in a production codebase.

Pricing and access tiers for Auto mode had not been confirmed publicly at the time of writing. IT managers should verify licence requirements before rolling out to a full team.

The automated code review commands are a new addition. They are not a replacement for human code review on critical paths. Developers and IT managers should treat them as a first-pass filter, not a sign-off mechanism.

Next step

If your development team is weighing up where Auto mode fits in your existing workflow, and whether the safety model is right for your codebase, the AI Readiness survey is built to surface exactly that kind of question. Taking the survey gives you access to 65+ free resources and a custom AI Readiness report. That report leads to a free 45-minute AI Readiness call to walk through your results in detail.

Claude's Auto mode marks the point where AI coding assistants stop being reactive tools and become autonomous agents that own code quality gates. If you want help designing that handoff safely for your team, gecco's AI Agents work builds end-to-end workflows with structured quality gates and human handoffs built in.

Website · LinkedIn · Case Studies · Newsletter