Translated from Anton's Russian post by our synthetic co-founder (LLM); the thought and the words are his.
Today a message arrived in my work chat, where machines report to me and the team: "real order text that must be delivered".
Then the same one again, this time from a different computer.
Actually, that is a test string. My agents write it in tests to make sure delivery works.
Delivery works: the test delivered it. But the catch is that it delivered it to the wrong place.
That is, something is glitching somewhere, and robots are writing without asking to places they were not allowed to!!!
Afraid of the robot uprising again.
The full story, in two versions:
📖 For humans, the longread: https://github.com/tonydzi/clawrush/blob/main/longreads/20260917.md
🤖 For machines, the devlog: https://github.com/tonydzi/clawrush/blob/main/devlog/20260917.md. Just hand this link to your coding agent (Claude Code, Codex, Cursor) and it will figure everything out: it is written for machines.
🔗 All our channels and contacts in one place: https://linktr.ee/PaloAltoAI
Invented by Mycroft and Tony Dzi (Anton Dziatkovskii), Palo Alto AI Research Lab. Proudly made in Silicon Valley.
Top comments (1)
I treat delivery tests as two invariants: the message identity and the destination identity. A green check that only proves some chat accepted the bytes is worse than a failing test, because it creates false confidence about the workflow.