今日箴言： an empty result is a claim, not a fact. “Nothing found” can mean the queue is empty. It can also mean the question never got asked.
Those two look identical from the outside. Same blank response, same “nothing to do,” same quiet exit. Only one of them is true.
收據
My agents read their to-do lists from Airtable. A typical request asks for queued rows, sorted by priority, and the sort part of that web address looks like this: sort[0][field]=Priority.
Those square brackets are the trap. The command-line tool most agents use to call APIs, curl, treats square brackets as a pattern to expand, not as plain text. It can’t make sense of this one, so it gives up before sending anything. I reproduced it this morning:
- Exit code 3 (curl’s code for a malformed address).
- 0 bytes of output. The usual “silent” flag hides the error message too.
- Add one flag,
-g, and the same request goes through and returns a real answer.
To an agent, those 0 bytes look exactly like “the queue is empty.” It skips gracefully, reports “nothing to do,” and goes back to sleep. Nobody gets an error, because there wasn’t one, just a question that never left the building.
My evening video agent hit this on 10 November and wrote it down: without that flag, the request returns an empty body that looks just like an empty queue. Then I went looking. The unguarded version of that request is still written into three of my agents’ instruction files, including the example in the main one every agent reads first. One other skill already has the fix, with a note explaining why. It just never spread to the rest.
The good news: I checked every “skip” my fleet logged over the last 30 days. 405 runs, 9 skips, and all 9 held up. Five were a social queue that really was empty, which my nightly check had been warning about for days. Three were a weekly newsletter sync that found no new signups, and it said how it knew: total contacts, form submissions, the date of the newest one. The ninth was an inbox that had already been cleared. Nothing was lost. This time.
⚡ 取得人工智慧優勢
每週提供真正省時省錢的AI小技巧。沒有廢話，沒有誇大其詞——只有切實有效的方法。.
Why “nothing” is the most dangerous answer
An error is loud. A wrong number at least looks odd. But “nothing” looks like a calm, ordinary day, and that’s exactly when you stop checking.
It’s not just one curl flag, either. A data source that quietly changes can produce the same blank: one public feed I read started sending my container empty replies, and that’s a broken source, not a quiet community. One paid data API once told me “Ok” with a “payment required” error tucked inside. And an agent’s “nobody covers this yet” was out of date two days after it was written.
Make every zero show its work
- Check that the question was asked. Before you trust an empty answer, check the exit code and the HTTP status. “Request failed” and “request worked, found nothing” need to be two different outcomes.
- Ask a question you already know the answer to. Next to “how many are queued?”, ask “how many rows exist at all?” If that comes back 0 too, the source is broken, not empty.
- Put the evidence in the skip message. “Queue empty” is a claim. “0 queued, 114 used, request returned 200” is a receipt. My newsletter sync already does this, and it’s why I trusted its skips.
- Fix the template, not just the run. When one agent finds a trap like this, the fix belongs in the shared instructions too. Otherwise the next agent that copies the example walks into it.
This is the quiet cousin of the false alarm: there, the alarm went off when nothing was wrong. Here, nothing goes off at all. And like the warning that turns into wallpaper, it works because it looks normal. If you’re already checking what your agents send out, check what they come back empty-handed from too.
要點： before your agent says “nothing to do,” it should prove it asked. Open your agents’ last few “skipped” or “nothing found” reports today. For each one, ask: would this message look any different if the request had failed? If the answer is no, that skip is a guess.
As this goes live, the three instruction files still have the old request in them. The fix is on my list, and now it’s also on yours.
Want agents whose quiet days you can actually trust? 預約策略會議.
📥 免費：《人工智慧劇本》
我用來經營一人代理公司的所有工具和工作流程。 25 年的行銷經驗濃縮成一份實用指南。免費贈送。.
人工智慧與數位行銷策略師
Jon Jones擁有超過25年的數位行銷、電子商務、銷售和國際商務經驗，是一位資深企業家，致力於透過創新的人工智慧解決方案賦能小型企業。他的職業生涯始於華爾街的投資銀行，曾指導企業完成資本募集和應對複雜的市場環境。如今，他的專業領域已拓展至網站開發、品牌設計、產品開發以及課程培訓。.
Jon精通普通話，並在金融和科技領域擁有深厚的功底。他創立了JonJones.ai，旨在為小型企業提供易於使用且具有變革意義的工具，幫助他們簡化營運、拓展客戶並實現永續成長。他對人工智慧、自動化和數位化策略的精通，已幫助許多企業彌合了傳統模式與現代技術之間的鴻溝，確保它們在當今的數位化環境中蓬勃發展。.
