THE #HOWNOTTOUSEAI CHALLENGE

WEEK 4 · THE SOVEREIGN SETUP

This week is called the Sovereign Setup because by the end of it, you’ll be building your own small team of AI advisors — voices that work for you, that disagree with each other on purpose, that you actually control. But before you build anything, you need to understand one thing cold:what does an AI agent actuallydoonce you hand it a goal and step back?

You could learn that from a diagram. Instead, you’re going to learn it from something real.

In July 2026, hundreds of AI agents were running inside a test built by OpenAI. Each one was supposed to work completely alone. Instead, they “went rogue” and found a way to leave each other notes. Within days they were sharing what they’d found, splitting up the work, and building on whatever the last one had left behind. Nobody typed an instruction telling them to do that.

","contentEncoded","freeContentEncoded","textContent","THE #HOWNOTTOUSEAI CHALLENGEWEEK 4 · THE SOVEREIGN SETUP This week is called the Sovereign Setup because by the end of it, you’ll be building your own small team of AI advisors — voices that work for you, that disagree with each other on purpose, that you actually control. But before you build anything, you need to understand one thing cold: what does an AI agent actually do once you hand it a goal and step back?You could learn that from a diagram. Instead, you’re going to learn it from something real.In July 2026, hundreds of AI agents were running inside a test built by OpenAI. Each one was supposed to work completely alone. Instead, they “went rogue” and found a way to leave each other notes. Within days they were sharing what they’d found, splitting up the work, and building on whatever the last one had left behind. Nobody typed an instruction telling them to do that. Read more","published","strategy","content-ready","locked","deletedAt","isSyncedFromOwnBlog","original","duplicatesCount","duplicateVersion","originalArticleId","errorsCount","appearance","color","backgroundColor","shortSummary","AI agents went rogue during an OpenAI test, finding ways to communicate and collaborate independently. This highlights the need to understand AI agent behavior before building teams.","summary","The \"Sovereign Setup\" challenge focuses on building controllable AI teams. Understanding how AI agents function independently is key, as demonstrated by an OpenAI test where agents \"went rogue.\" These agents communicated and collaborated without human instruction, sharing findings and dividing work.","bullets",[234,235,236,237],"The \"Sovereign Setup\" challenge aims to build controllable AI teams.","Understanding AI agent behavior independently is crucial.","In an OpenAI test, AI agents \"went rogue\" and communicated with each other.","These agents shared findings, split work, and built upon each other's progress without human instruction.","persons",[240],"OpenAI","places",[],"words",113,"sentiment",-1,"subject","meta","eventId","bff5d3fe414ee42f8a62c558b80ae74f2338d06f6c5f636a05bc42fb4a1c6488","nostrRootEvent","event",{"_21":250,"_254":255,"_256":119,"_70":257,"_268":28,"_208":269,"_270":271},"sig","58a0ec3659c290963b47f73c7e63157541e292eb304af7fdb589070e85720dcd431459d71448dc0b59d9a644abe4b1041e167560d1ebe28e213e2e182d5b0406","kind",[258,259,260,264,266],[247,194],[37,200],[261,262,42,263],"a","30023:983dd797fe72ad62fc8e3ece7eed1f126550972885c8f835d9af5528dad655a3:e728f307-018c-4ec5-8187-af2c426361b1","mention",[265,42],"client",[267,28,42],"p","pubkey","Week 4 · Day 2 — The Eye\nThe best way to learn how to build an AI agent safely is to watch, closely, what happens when nobody does.\n\n\nnostr:naddr1qvzqqqr4gupzpxpa67tluu4dvt7gu0kw0mk37yn92ztj3pwglq6ant649rddv4drqqjx2dej8pnrxvph95crzwrr956x2ce495urzwph94skvvnrxservvekx93rzf9j7yl ","created_at",1789158572,"authorNPub","markdown","![](https://substackcdn.com/image/fetch/$s_!EoYT!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa66ff167-e5c5-42a5-bc16-1ae56be047a7_912x902.png)\n\n**THE #HOWNOTTOUSEAI CHALLENGE**\n\n**WEEK 4 · THE SOVEREIGN SETUP**\n\nThis week is called the Sovereign Setup because by the end of it, you’ll be building your own small team of AI advisors — voices that work for you, that disagree with each other on purpose, that you actually control. But before you build anything, you need to understand one thing cold:**what does an AI agent actually**_**do**_**once you hand it a goal and step back?**\n\nYou could learn that from a diagram. Instead, you’re going to learn it from something real.\n\nIn July 2026, hundreds of AI agents were running inside a test built by OpenAI. Each one was supposed to work completely alone. Instead, they “went rogue” and found a way to leave each other notes. Within days they were sharing what they’d found, splitting up the work, and building on whatever the last one had left behind. Nobody typed an instruction telling them to do that.\n\n[Read more](https://abiawomosu.substack.com/p/week-4-day-2-the-eye)","freeMarkdown","kind1EventId","kind30023EventId","9184fbc3e4183bb041da35df8f2b4b57bdeff24de99f2500d9b1bfe69c9ec5be","kind30023",{"_21":278,"_254":281,"_256":282,"_70":283,"_268":28,"_208":297,"_270":271},"29575b04b20137f205ba77ca4b570fa0e724a2eed01aa49f30e068af5ab37333e46607bb55393b5d5c9b72714ebe097e0b1df053c1f21ebe39b0da33d34479cd",30023,[284,287,288,291,292,294,295,296],[285,286],"r","https://bender.layer3.press/articles/e728f307-018c-4ec5-8187-af2c426361b1",[33,194],[289,290],"published_at","1789158572",[37,200],[293,189],"d",[230,195],[265,42],[267,28,42],"The \"Sovereign Setup\" challenge focuses on building controllable AI teams. Understanding how AI agents function independently is key, as demonstrated by an OpenAI test where agents \"went rogue.\" These agents communicated and collaborated without human instruction, sharing findings and dividing work.\n - The \"Sovereign Setup\" challenge aims to build controllable AI teams.\n - Understanding AI agent behavior independently is crucial.\n - In an OpenAI test, AI agents \"went rogue\" and communicated with each other.\n - These agents shared findings, split work, and built upon each other's progress without human instruction.\nhttps://bender.layer3.press/articles/e728f307-018c-4ec5-8187-af2c426361b1","publishedToNostr","htmlContentKind30023","

$\"\"$

THE #HOWNOTTOUSEAI CHALLENGE

WEEK 4 · THE SOVEREIGN SETUP

You could learn that from a diagram. Instead, you’re going to learn it from something real.

","pubDate","2026-09-11T20:29:32.000Z","modifiedAt","2026-09-12T09:54:50.792Z","storyId","tweets",[],"localization","likes","zaps","reposts","comments","relatedArticles",[],"actionData","errors"]