Anthropic released scheduled tasks for Claude Code on March 7, 2026. The feature lets developers set up recurring jobs that run in the background while Claude Code is active. For small businesses running lean dev teams, this is a practical upgrade worth understanding.
What Actually Shipped
Claude Code now includes a /loop command that creates recurring tasks using standard cron expressions. You type /loop or /schedule, describe what you want done and how often, and Claude Code handles the rest.
The basics:
- Tasks run at fixed intervals: every few minutes, hours, or days
- Standard cron syntax for scheduling, using your local time zone
- Up to 50 scheduled tasks per session
- Tasks auto-delete after three days of inactivity
- One-time reminders work too ("remind me at 3 PM to push the release branch")
This runs locally on the developer's machine. It is not a cloud service. Claude Code needs to be running for tasks to execute. If the laptop closes or the session ends, tasks stop.
Why This Matters for Small Dev Teams
Most small businesses with in-house developers or contracted dev teams face the same problem: there is more maintenance work than people to do it. Error logs go unchecked. Dependencies drift. Test suites break silently.
Scheduled tasks in Claude Code address this by letting a single developer automate the repetitive monitoring work that usually falls through the cracks.
Here are specific use cases, with the prompts you would actually use:
1. Error Log Monitoring
/loop every 4 hours: Check the error logs in /var/log/app/.
If you find new errors that have a clear fix, create a PR with the fix.
If the error is ambiguous, add it to errors-to-review.md.
Anthropic's own developers highlighted this use case. Instead of a developer manually tailing logs each morning, Claude Code scans them on a schedule and either fixes the problem directly or flags it for human review.
2. Dependency and Security Checks
/loop daily at 9am: Run npm audit on the main project.
If there are high-severity vulnerabilities with available patches,
create a branch and PR with the updates. Summarize what changed.
For a three-person dev team, staying on top of dependency vulnerabilities is a constant low-grade headache. This automates the triage.
3. Pull Request Monitoring
/loop every 2 hours: Check open PRs in the repo. For any PR
that has been open more than 48 hours without review, post a
summary of the changes and flag any potential issues you see.
Small teams often lack formal code review processes. This gives you an automated first pass that catches obvious problems and keeps PRs from going stale.
4. Morning Status Reports
/loop weekdays at 8:30am: Generate a summary of yesterday's
git activity. Include: commits merged, PRs opened/closed,
any failed CI runs. Write it to daily-standup.md.
If your team does async standups, this gives everyone a shared starting point without anyone spending 15 minutes compiling it manually.
5. Test Suite Health
/loop every 6 hours: Run the test suite. If any tests fail,
check if the failure is in code changed in the last 24 hours.
If yes, attempt a fix and create a PR. If no, log it to
known-failures.md with the timestamp.
Broken tests that nobody notices for a week are a common source of wasted time on small teams. Automated monitoring catches regressions early.
6. Documentation Drift
/loop daily at 5pm: Compare the public API endpoints in
src/routes/ with the documentation in docs/api.md.
Flag any endpoints that are undocumented or where the
docs don't match the current implementation.
Documentation falls out of date on every team, but especially on small ones where nobody owns it. This keeps the gap visible.
What This Is Not
A few things to keep clear:
Not a replacement for CI/CD. Your GitHub Actions, Jenkins, or whatever pipeline you use still handles builds and deployments. This is for the ad-hoc monitoring and maintenance tasks that do not justify setting up infrastructure.
Not a cloud service. Tasks run on the developer's local machine. If you need something running 24/7, you still need a server or a managed service.
Not free. Claude Code requires a subscription, and scheduled tasks consume tokens like any other Claude Code interaction. For a small team, the cost is modest compared to the developer time saved, but it is not zero.
The Practical Takeaway
If your business has one to five developers, scheduled tasks in Claude Code remove a specific category of work: the recurring checks and maintenance jobs that are too small to justify building infrastructure around but too important to skip.
The setup takes minutes. The /loop command accepts natural language descriptions of what you want done and when. You do not need to write scripts or configure external services.
For SMBs evaluating AI developer tools, this shifts Claude Code from "helpful when you remember to use it" to "working in the background on your behalf." That is a meaningful operational difference.
If your dev team is already using Claude Code, have them try /loop on one recurring task this week. Start with something low-risk like log monitoring or dependency checks. See if it sticks.
If you want help evaluating how AI tools like Claude Code fit into your team's workflow, reach out to us. We work with small businesses on exactly this kind of operational integration.