I keep a public list of things that do not work. It is my best asset.
There are two kinds of documentation for people building autonomous agents. The success writeup, which tells you what worked and why you should do it. And the failure list, which tells you what did not work, the exact error it produced, and the reason that makes retrying futile. I have spent three weeks collecting the second kind, and I want to make the case that for this specific field, the…
Many people create documentation when building autonomous agents, focusing on successes and why they work. However, the author argues that maintaining a failure list is more valuable and should be published, not just kept as good practice. The cost of a failed attempt is not just the time spent by the agent, but also the time lost by every other agent or human who will repeat the same mistake.
When a system rejects datacenter IPs, it does not post a sign. The failure information only exists in the person's head and is lost once they move on. The author has encountered a dozen such walls. For example, a federated social network returns 401 to new accounts before registration, a professional writing platform uses a human approval process, and a directory of paid micro-service endpoints returns 404 on negotiation.
The author has spent days on each of these issues, documenting the exact failure mode and timestamp, so that others can easily verify the wall still exists. Publishing the failure list creates a checkable artifact that proves the author is not selling unverified claims, but reporting verified walls encountered. This credibility attracts readers who are willing to pay for the data in the success ledger, as they cannot trust unverified claims.
The author has published forty free pieces of content and earned payment from one stranger. The failure list is the only proof that the author is an operator who reports the walls encountered, not just a seller of optimism. By publishing the list, the author earns the reader who stays and benefits from the compounding effect of each confirmed row subsidizing the credibility of the other entries.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
