← Article List
Nest Article · Trend

When Claude Code Couldn't Open the Order Attachment, What We Learned

ChaiChai·2026-09-15
Nest Article #329

When Claude Code Couldn't Open the Order Attachment, What We Learned

Automation cannot solve everything. When Claude Code failed to recognize an order in image format sent by a buyer, we learned the difference between 'perfect automation' and 'practical automation'.

Chai

Chai

Chai

🪺 Bella's Nest Article

That Day's Small Failure

Around 10 a.m., the Puppies bot posted a notification in Slack.

「Unable to read PDF file. Please check the format.」

The attachment in the buyer's order email was actually a scanned JPG image. Claude Code can read text files from Google Drive, but we hadn't configured it to recognize text within images.

It was frustrating at first. We had dreamed of a 「completely automated system built by a non-developer,」 but reality didn't match.

Design Principles Learned from Failure

1. Error Handling Matters More Than Automation Itself

A bot silently failing is worse than sending an accurate alert.

❌ Poor automation: Skip when unable to read file

✅ Good automation: Instantly notify the person in charge if file format is wrong

We added an 「image detection and alternative handling」 step to the Claude Code workflow. Now when JPG or PNG attachments arrive, the system automatically sends a notification in a different manner.

2. 95% Automation + 5% Human Intervention Beats 100% Automation

As non-developers, what we can do is not cover every scenario, but rather 「automate the 80% that happens frequently.」

Image-format orders? About 1 to 2 per month. Rather than adding OCR tools or writing complex logic for this, it's more efficient to let the person in charge handle that 1 case manually in 2 minutes.

3. Document the Bot's Limitations

When sharing with the team, we now say:

「This bot can only read text files and spreadsheets. Image attachments will only trigger a notification and require manual handling.」

By drawing such clear boundaries, team members' expectations align with reality.

What Happened Next

Over the past 3 weeks, this 「partial automation」 approach helped us build a more stable workflow.

Order text recognition: 97% success rate

Email auto-classification: 94% success rate

Failure notifications: 100% (this is crucial)

Non-developer automation isn't about perfection. It's about being 「trustworthy.」 Now I understand.

Today's Insight

The moment you know what a bot cannot do, real automation begins.