9 AM, Slack Message
I got a notification that yesterday's scheduled overnight automation task didn't run. For the first time, our dream team bot (built with Claude Code across Windows and Mac mini environments) simply declined to execute.
Instead of "Why didn't it run?", my first thought was "Is this the right signal?"
The Core Issue
I had set up the bot to run data validation and multilingual content generation on Google Sheets during overnight hours (2 AM to 5 AM). It hit Claude Code's runtime resource limits.
Constraints:
• Continuous execution time limit (roughly 30 minutes)
• Async job scheduling not supported
• Memory accumulation issues (happen during long runs)
As a non-developer, I initially thought "the computer can't do it." What it really meant was "it can't do it this way."
Solution: Redesigning Automation
Step 1: Task Chunking
I broke the massive overnight job into 3 smaller tasks, each completing within 10 minutes.
# Before: Single integrated task (30+ minutes)
process_all_data_and_generate_content()
# After: 10-minute unit tasks
step_1_validate_sheet_data() # 5 min
step_2_generate_descriptions() # 7 min
step_3_update_translations() # 6 min
Step 2: External Scheduler Integration
Used Windows Task Scheduler to run the 3 jobs sequentially.
• 2:10 AM: Data validation
• 2:25 AM: Content generation
• 2:40 AM: Multilingual translation update
Step 3: Failure Recovery Logic
Each step automatically rolls back if it fails. I chose "safe automation" over "perfect automation."
try:
step_2_generate_descriptions()
except Exception as e:
rollback_to_step_1()
send_slack_alert("Step 2 failed, rolled back to Step 1")
Lessons Learned
The Real Limit of Automation Isn't Technology, It's Design
I initially blamed Claude Code for being insufficient. The truth was I tried to do too much in one shot.
Running 24/7 Doesn't Make a Bot Good
Concentrated overnight processing and distributed all-day processing each have trade-offs. Since our team works better with overnight batches, we should design around that.
Automation Without Monitoring Is Just Neglect
We discovered the failure late because we weren't checking logs daily. Now we have real-time Slack alerts.
Next Steps
I'm creating a simple template for other non-developers facing similar challenges. Planning to share a "5-Point Checklist for Safe Long-Running Automation" with the dream team.
The bot's refusal wasn't failure. It was the system's message: "Design smarter."