AIToday
Large Language ModelsAI Coding AssistantsZenn AI/MLPublished: Nov 16, 2026, 16:00 JST

Solo developer runs 15 Claude Code departments, 48 jobs

Send on LINE
Solo developer runs 15 Claude Code departments, 48 jobs

3 Key Points

  1. What happened

    A freelance web and SEO tool developer reports running Claude Code with one COO instance, 15 sub-agent departments defined in .claude/agents/, and 48 loaded launchd jobs as of Nov 13, 2026.

  2. Why it matters

    The setup shows how a single operator can delegate recurring work across departments, even as the body notes most of this scaffolding needs to be found by search rather than built in advance.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

The setup is a response to a specific failure mode: scheduled Claude Code jobs that stop running without telling anyone. The developer's fix is to move reporting out of the AI itself, using a shell-side trap so a message arrives even when Claude crashes or times out. A separate nightly watcher then looks for jobs that never loaded, resident processes that fell over, and reports more than three hours late — with a carve-out for times when the Mac was asleep so those aren't counted as anomalies.

The body also describes a learning ledger: decision-making jobs read a summary of prior results, record one line per outcome, and propose changes to the human rather than rewriting procedures automatically, on the reasoning that wrong lessons would quietly pile up. Departments were added only after a visible bottleneck, such as a quiz tester split off because the problem creator shared the same assumptions as the reviewer. Three recent failures — a long UTM URL posted to social media, a sales job that sent zero messages twice, and an accidental push of 11 commits — each added one rule or hook. The developer's stated starting order is to build one read-only scheduled job with a completion report, then a watcher, before adding more departments.

FAQ
How many scheduled jobs are actually running?
As of Nov 13, 2026, 48 jobs are loaded, out of 60 registered; the other 12 are stopped, each with a reason recorded in a config file.
How does the developer know if a scheduled job failed?
A shell-side trap sends a report email on success, failure, or timeout, and a separate watcher job checks for jobs that never loaded or missed their report deadline.
What stops the AI from making dangerous changes on its own?
PreToolUse hooks route risky commands like rm -rf or git push --force to human confirmation, and a pre-push check blocks commits containing keys or .env data.
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleThailand GDP grows 2.9% on AI-related exports