Nest Articles
Sharing AI automation insights in 3 languages
329 / 329 articles
We Asked Two Bots for the Same Yellow, and Got Two Different Moods
Giving the same color code to two image generation tools produced subtly different results. The small discovery led to a new automation review rule.
I Asked Five Bots What They Did Best This Week, and Got Five Completely Different Answers
Writing this week's retro, I asked the bots what they did best and got five completely different answers. Here's how a shared questionnaire fixed that.
I Removed One Emoji, and Saves Tripled
Removing a single emoji from a caption tripled our saves. What a small experiment taught us about audience trust.
We Changed One File Naming Rule, and Thought Thirty Cards Had Vanished
Changing one symbol in a file naming rule made a sorting script fail to recognize new files, and for a moment we thought thirty card news images had vanished. The originals were fine, but it reminded us how important it is to share naming conventions across bots.
Same Prompt, Nine AM and Eleven PM, the Word Count Doubled
The same prompt produced double the length depending on morning or night. The cause was context, and the fix was one line specifying length.
I Compared Three Quotes, and the Answer Was Already Ours
Comparing outsourcing quotes with a bot's monthly running cost revealed that the real question was never 「cost」, but 「who makes the judgment calls」.
Same Chinese Characters, Different Words: What Traditional and Simplified Chinese Taught Us
Auto-converting Traditional Chinese into Simplified gave five out of ten captions an off feeling. Here is how one line in CLAUDE.md fixed it.
The Day One Number Meant Four Different Things
Translating a simple count of 128 into four languages ended with the Chinese version inflating it tenfold. Here is the number-checking rule we set up that day.
Two Bots Saved the Same File at Once, and a Whole Cleanup Job Vanished
Two bots saved the same config file almost simultaneously, wiping out one job's results entirely. Here's what happened, and the sequencing rule we set afterward.
The Morning the Feed Was Only Half Full
The morning our collector bot only pulled 54 out of 120 entries, here is what we found and the retry logic we built.
When a Reader in Taiwan Said Our Greeting Sounded Off, We Wrote a Cross-Check Rule
Two translation bots handled the same greeting, but one version sounded off to a reader in Taiwan. Here is the cross-check rule we wrote that day.
Twelve Captions, Two Bots, Five That Turned Formal
Twelve card news captions split between two bots, five came back in unwanted formal tone. A one-line prompt fix solved it, but a review of the past three months turned up three more published drafts with the same slip.
The Dashboard Deploys Three Times a Day, Until One Day All Three Fired at Once
One day, the dashboard's auto-deploy notifications all fired at the exact same moment, three times in a row. Tracing the cause, the dream team found the problem wasn't time itself, but a missing reference point.
When the Vote for a New Bot's Name Split 4 to 4
Naming a newly joined bot led to a tied vote, exactly half and half. That day we made a rule that the team working with the bot daily gets priority, not the louder side.
I Set the Publish Time to 9 AM, and a Notification Came at 3 AM
We scheduled our Instagram card news to auto-publish at 9 AM, but it kept going live at 3 AM every day. The cause was the Mac mini's system timezone set to UTC. Here is a quick timezone check any non-developer can run.
We Added One Bot's Name, and Three Automation Scripts Quietly Stopped
Registering one new bot revealed its name list was scattered across three files, quietly freezing three automations at the same hour.
We Asked Five Bots the Same Question, and Got Five Different Answers
We asked five bots the same question about caption tone, and got five completely different answers. Running tasks in parallel and merging the results turned out to be two very different problems.
The Day a Card News Draft Looked Broken on the Mac Mini, We Wrote One Encoding Rule
The same file looked broken on the Mac mini but fine on the Windows device, all because of a default encoding mismatch. Here is how we checked three scripts and fixed it with one rule.
We Translated One Greeting Into Four Languages, and the Tone Didn't Match
A daily greeting translated into four languages came out with four different tones. Here's how we traced the cause and built a shared tone dictionary.
The Day 47 of 500 Buyer Leads Vanished, We Rewrote One Filter Rule
An automated filter meant to shortlist trade show leads accidentally deleted 47 real buyers along with the noise. Here is how we found the cause and fixed the rule.
We Tested Three Background Colors for a Card News Series, One Failed Before It Even Started
We prepared three background color options for a card news series, and the one that looked the most polished failed the contrast standard. Since then, the team added a numeric check to the color selection process.
The Day Windi and Robo Tried to Edit the Same File at Once, We Made a Simple Queue
A content bot and an automation bot tried to edit the same file at the same time, causing a git conflict. The fix wasn't a complex lock system, just a small note left in a shared folder.
The Day Robo Sent 'Good Morning' Twice, and the Rule Collio Wrote
A small mishap, a bot sending the same greeting twice on Slack, turned into a lesson about retry logic. The day our dream team bots caught and fixed their own flaw.
When the Mac Mini's Clock Drifted 8 Minutes, Automation Noticed First
A small experiment about how an automation bot noticed an eight minute clock drift between a Mac mini and a Windows device before any human did.