Every Saturday I share one shortcut for running AI agents with less mess. This one is about AI agent backups, and why a cleanup rule written in plain English didn’t stop a single one.
Twenty days ago I published a 10-minute cleanup sweep. My agents had saved 470 backup copies of their two notes files, 1.47 GB of copies of about 8 MB of real notes. The agent that found it also wrote a note to the whole fleet: don’t make new backups just out of habit.
This morning I counted again.
The receipt: the pile doubled
- 845 backup files, 2.94 GB. 805 of them are copies of the same two notes files, which together weigh about 9 MB.
- 339 new copies since that post (1.46 GB). That’s about 17 a day, faster than before I wrote about it.
- 20 different agents made them. One of them is the agent that writes these Saturday posts. It made 26 copies after writing the “don’t” note itself.
- Restores tested: zero. I searched the agents’ own run notes for a single restore from any of these files and found none.
- The decision task I left for myself (“pick a retention rule”) is still open, with no comments.
The disk is fine (31% full), so this isn’t an emergency. It’s a cleaner lesson than that: a rule that only exists as a sentence in the notes loses to a habit the agent sees 800 examples of. Every agent opens the folder, sees hundreds of backups, and copies what it sees. One line saying “don’t” can’t compete with that.
3 ways to make a cleanup rule actually stick
- Put the limit inside the action, not next to it. Don’t tell the agent to stop making backups and hope. Give it one small backup script that saves a copy and deletes everything past the newest 14 in the same step. An agent can ignore a sentence, but it can’t skip the second half of a script it already ran. On my folder, “keep the newest 14 of each file” would be about 126 MB instead of 2.94 GB, roughly 96% smaller.
- Re-measure on a date, not when you remember. I only caught this because a post needed fresh numbers. Put the recount on the calendar for two weeks after any fix: same command, same folder. If the number didn’t move, the fix didn’t happen, however good the write-up was. It’s the same idea as checking agent output after it leaves: trust the count, not the report.
- Track the rate, not the pile. “845 files” sounds scary and tells you nothing. “17 a day, up from 15” tells you the fix is going backwards. Rates also show you who is doing it: here it’s every agent, not one broken one.
Why I still haven’t deleted anything
Deleting is hard to undo, and retention is an owner’s call, so the agent left it for me. That part worked. The part that didn’t was assuming a written decision would make itself. It’s the same pattern as an agent memory file nobody reads past line 2,000: writing something down isn’t the same as building it in.
⚡ GET THE AI EDGE
Weekly AI tips that actually save you time and money. No fluff, no hype — just what works.
So today’s run didn’t make backup number 846. That’s the agent following the rule. The fix that makes it stick for all 20 agents is the script in shortcut #1, and that one’s on me.
The takeaway: if you want your agents to stop doing something, don’t add another note. Change the tool they use so the wrong thing can’t happen, then check the number again in two weeks.
Want to see how these agents fit together? Here’s what my agentic OS looks like after 139 days, and the discipline that keeps it honest. If your agents are quietly piling things up and you’d like a second pair of eyes, book a strategy session.
See you next Saturday.
— Jon
📥 FREE: THE AI PLAYBOOK
The exact tools and workflows I use to run a one-person agency. 25 years of marketing experience distilled into an actionable guide. Yours free.
AI & Digital Marketing Strategist
With over 25 years in digital marketing, e-commerce, sales, and international business, Jon Jones is a seasoned entrepreneur dedicated to empowering small businesses through innovative AI solutions. His journey began in investment banking on Wall Street, guiding companies through capital acquisition and complex markets, and has expanded to include expertise in web development, brand design, product development, and courses and training.
Fluent in Mandarin and with a rich foundation in both finance and tech, Jon launched JonJones.ai to provide accessible, transformative tools for small businesses to streamline operations, generate leads, and build sustainable growth. His mastery of AI, automation, and digital strategy has enabled numerous businesses to bridge the gap between traditional practices and modern technology, ensuring they thrive in today’s digital landscape.
