Why Your AI-Generated Automation Needs a Human to Maintain It
The Maintenance Myth
There’s a popular belief that once an automation is built, it runs forever. Set it and forget it. Especially if AI wrote it; it’s code, it doesn’t degrade, it doesn’t get tired.
This is true of the code itself. A script you deploy today will run the same instructions a year from now, character for character.
But the environment it runs in? That changes constantly. And when the environment shifts, even perfect code stops working.
What Changes Around Your Script
Google Updates Their APIs
Google regularly updates, deprecates, and modifies their Apps Script services. Sometimes these changes are announced; sometimes they’re subtle behavioural shifts that only surface when your script suddenly does something different.
Recent examples:
- Changes to how
GmailApphandles label permissions - Modifications to
DriveAppbehaviour with Shared Drives - Updates to
SpreadsheetApp.flush()timing - Changes to OAuth scope requirements
When Google changes something, your AI-generated script doesn’t know. It keeps running the old way; which might now be the wrong way.
Your Data Changes Shape
Your script was built against today’s data structure. But businesses evolve:
- Someone adds a column to the spreadsheet and all your column index references shift by one
- A form gets a new question and the response mapping changes
- Client email formats change when their company rebrands
- File naming conventions evolve and your pattern matching stops matching
- Data volumes grow from 100 rows to 10,000 rows and the script times out
AI built the script to match a snapshot of your data. Data doesn’t stay in snapshots.
Your Business Process Changes
The workflow you automated in January might not be the workflow you need in July:
- New approval steps get added
- Client communication templates change
- Reporting requirements expand
- Team responsibilities shift
- Regulatory requirements update
Each of these changes needs the automation to update too. And unless someone understands the code well enough to modify it safely, the automation becomes a rigid constraint instead of a flexible tool.
Permissions and Access Change
Google Workspace permissions are dynamic:
- Team members join and leave
- Shared Drive access gets reorganised
- OAuth tokens expire or get revoked
- Admin policies change what scripts can do
- Third-party app access reviews block your script’s connections
AI doesn’t configure scripts for permission resilience. When access changes, the script fails; often silently.
The Three Types of Maintenance
1. Reactive: Something Broke
The script stopped sending emails. The spreadsheet hasn’t updated since Tuesday. A client didn’t receive their report.
Reactive maintenance is the most expensive kind. By the time you notice, damage has been done; missed communications, inconsistent data, broken processes. The fix itself might be quick, but finding the cause takes time, and repairing the downstream effects takes more.
AI-generated scripts are especially vulnerable to reactive maintenance because they typically lack monitoring. There’s no alert when something fails. You find out when a human notices the absence of something that should have happened.
2. Preventive: Keeping Things Running
Regular checks that catch problems before they become incidents:
- Verifying triggers are still active
- Checking execution logs for errors or warnings
- Monitoring processing times (gradual slowdowns often predict failures)
- Testing after Google Workspace updates
- Reviewing quota usage trends
This is the maintenance that AI never builds and rarely suggests. It’s not glamorous, but it’s the difference between “the automation works” and “the automation reliably works.”
3. Adaptive: Evolving With the Business
The most valuable maintenance isn’t fixing broken things; it’s adapting working things to changing needs:
- Adding new fields to process when forms are updated
- Adjusting logic when business rules change
- Optimising performance when data volumes grow
- Extending functionality when the team asks “can it also do X?”
This requires understanding both the code and the business context. AI wrote the code, but it doesn’t understand your business. When you need the automation to evolve, you need a human who understands both.
What Good Maintenance Looks Like
Built-In Monitoring
Every production script should include:
- Execution logging: A record of every run; timestamp, items processed, errors encountered
- Heartbeat checks: A separate script that verifies the main automation ran successfully
- Alert system: Email or Slack notification when something fails or when a metric exceeds normal range
- Dashboard: A spreadsheet tab or simple web view showing automation health at a glance
This isn’t extra work; it’s part of the build. The monitoring often takes 20% of the development time but prevents 80% of maintenance headaches.
Documentation
The script needs documentation that a human can read:
- What does this script do (in plain English)?
- What triggers it?
- What data does it read and write?
- What are the known edge cases and limitations?
- What should you check if it stops working?
- How do you update the email template / keywords / thresholds?
AI rarely documents its code in a way that’s useful for ongoing maintenance. Comments describe what the code does (which you can see) but not why it does it or what to do when it breaks.
Regular Review Schedule
- Weekly: Check execution logs for errors
- Monthly: Review processing volumes and performance
- Quarterly: Verify the automation still matches the business process
- After any Google Workspace update: Test critical automations
The Cost of No Maintenance
I regularly take on clients whose AI-generated or DIY automations have been “running” for 6-12 months with no maintenance. The common findings:
- Triggers that stopped firing weeks ago (nobody noticed)
- Error logs full of repeated failures (nobody checked)
- Data inconsistencies from silent processing errors
- Outdated logic that no longer matches the current business process
- Duplicate triggers sending multiple copies of the same email
- Scripts hitting quota limits during high-volume periods
The recovery cost is always higher than the maintenance would have been. Sometimes significantly higher, because data inconsistencies need to be traced back and corrected manually.
The Human in the Loop
AI can write your automation. It can even write decent automation. But it can’t:
- Monitor the automation after deployment
- Understand when your business process changes
- Adapt the code to new requirements
- Debug failures that come from environmental changes
- Make judgement calls about when to alert and when to auto-recover
These aren’t AI limitations that will be solved with the next model. They’re fundamental to the nature of ongoing operations in a changing environment. The automation runs in a world that changes daily; someone needs to make sure it keeps up.
Want Automation That Comes With Maintenance?
At Empower Automation, we don’t just build and walk away. Every automation we deploy includes monitoring, documentation, and a maintenance plan. We offer ongoing support packages for businesses that want someone keeping an eye on things; so you can forget your automation exists and trust that it’s still working.
Book a free 15-minute automation audit →
If you’ve got automation that was built by AI or a developer who’s since moved on, we’ll audit it, tell you what state it’s in, and recommend what it needs to keep running reliably.
Nicola Berry is the founder of Empower Automation, based in Falkirk, Scotland. Building automation that lasts, and maintaining it when the world changes.
Professional Email & Tools for Your Business
Get you@yourcompany.com, plus video meetings, secure cloud storage, and the powerful admin controls you need to scale. Same tools I use to build your automations.