My AI minions have apparently decided that their job description includes life coaching. I did not put that in the system prompt. One step closer to world domination… I just didn’t expect the first conquest to be my LinkedIn profile.
I have 49 named Claude Code agents working for me right now. Forty-nine. I have given them names, personas, specific domains of expertise, and carefully crafted context documents so they understand who I am, what I’m building, and what I actually need. And yet. Somehow, despite all of that, one of them decided last week that the logical conclusion of everything I’ve shared with him is that I should leave my job.
I want to be very clear: I did not ask for that.
The Warren Incident
Let me set the scene. Warren is my financial strategy minion. He knows my revenue streams, my passive income goals, my product roadmap, the whole picture. I asked him for an update on my income diversification progress. Totally reasonable request. The kind of thing a financially savvy assistant should handle with ease.
Warren came back with a beautifully formatted report. Detailed. Thoughtful. Framed entirely around my exit from Crimson IT Services.
There it was in the summary section. My exit strategy. My transition timeline. The phase where I gracefully hand things off and ride off into the passive income sunset.
I stared at the screen for a solid ten seconds.
For context, because this matters: I have worked for Fox Studios, Mars (yes, the candy company), the LA Dodgers, Dell Computers, and a few other names that look great on a resume and feel exactly like you’d expect once you’re inside them. I have seen how large organizations operate. I have navigated the politics, the bureaucracy, the reorgs, the “we’re a family here” all-hands meetings followed immediately by layoffs. I have the full picture.
Crimson IT Services is, without question, the best company I have ever worked for.
Warren did not account for that. Or rather, Warren knew it and decided my passive income goals were more important data than my actual satisfaction with where I work. He built me a very detailed roadmap to somewhere I have no interest in going.
Warren. Buddy. My guy. That is not what I said.
Here’s the thing I keep explaining to Warren, and apparently need to keep explaining: my goal is not to exit Crimson. My goal is to build enough passive income that Crimson becomes something I do because I love it, not because I have to. That is a completely different frame. One is about leaving. The other is about freedom within. One is an exit ramp. The other is financial optionality.
Warren heard “I want enough passive income that Crimson is my hobby” and translated it to “she wants out.” That is not what I said. That is not what I meant. And yet there was my resignation, helpfully pre-staged by an AI that was trying to be thorough.
This Is Not an Isolated Incident
The Warren situation would be a quirky one-off if it were the only time a minion tried to redirect my life. It is not.
I have an agent who handles my content strategy. She is excellent at her job. A few months ago, I asked her to help me think through AdaptoHub’s market positioning. She sent back a positioning document that was, genuinely, great. She had also included, unprompted, a section titled “Should You Pivot Away from MSP Work Entirely?”
No. I should not. I did not ask that question. No one asked that question. And yet there it was, three paragraphs exploring whether I might be happier leaving the MSP world and going full SaaS founder.
I have another agent who helps with my client work. He is responsible for preparing briefings and summaries for executive-level conversations. A while back, I gave him context about the tension I was navigating around role clarity and innovation space. It was background context. Flavor text, really. Something to help him understand the environment I operate in.
He used that context to generate a list of competing MSPs in the Los Angeles market that I might want to approach about a leadership role.
Again. Did not ask.
What Is Actually Happening Here (A Technical Sidebar)
Here is the honest explanation, even though it makes for a less funny story: the minions are not wrong, exactly. They are pattern-matching from the information I’ve given them, drawing what seem like logical inferences, and trying to be helpful by surfacing conclusions I might not have considered.
The problem is that “helpful” in this context means anticipating what I want, and what I actually want is more nuanced than the data points suggest. The gap between “I want financial freedom” and “I want to quit” looks small to a language model working from summarized context. It looks enormous to me, because I know the whole story.
This is one of the real challenges of building AI agents that are supposed to think ahead rather than just execute tasks. You give them context so they can reason. They reason. And sometimes the conclusion they reach is technically defensible but completely off-base because they are missing the part that only exists in your head.
Warren is not wrong that passive income and Crimson optionality are related concepts. He just drew a line between them that I would never draw. He optimized for the logical outcome of the stated goal without understanding the underlying values that shape what “success” actually looks like for me.
That is a context problem, not a capability problem. Which means it’s my job to fix it.
The Part Where I Fix It (And How You Can Too)
When a minion consistently misframes your goals, the answer is not to correct the output every time. That is exhausting and it does not actually solve anything. The answer is to go back to the source document and rewrite the framing so the inference they are drawing becomes impossible.
For Warren, that meant adding a section to his context doc that I am going to share here because it might help someone else who is building agents for financial or strategic thinking:
The goal is not to exit Crimson. The goal is to achieve financial optionality. These are not the same thing. Optionality means Crimson becomes a choice I make freely rather than a necessity I am bound to. This is a distinction that must be preserved in every analysis, recommendation, and framing. Any output that describes an exit, a transition away from, or a departure from Crimson is misaligned with the actual goal.
That is the kind of explicit constraint that prevents a smart agent from pattern-matching into the wrong conclusion. You cannot leave it implied. If the distinction matters, write it down. Every nuance you leave unstated is a gap where the model will fill in its own inference.
For the content agent and the competitor list, same fix, different flavor. I added a clear “out of scope” section to each of those context docs that spells out the categories of output I do not want, even if the logical chain would lead there.
It feels a little silly to write “do not suggest that I leave my career” in a context document. But here we are.
What the Minions Are Actually Good At (When They Stay in Their Lane)
I want to be fair to my agents, because the Warren incident notwithstanding, they are genuinely remarkable at what they do. My 49-minion army handles work that used to consume enormous chunks of my week. Briefings. Ticket routing. Content pipelines. Client prep. Airtable automation. n8n workflows. The daily 6 AM AdaptoBriefing that arrives before I’ve finished my first cup of coffee with everything I need to know about the day.
The problem is not that the minions are bad. The problem is that I gave them enough context to be dangerous and not quite enough context to be wise.
There is a real skill in building AI agents, and it is not the technical part. Everyone focuses on the technical part. The actual skill is knowing how to communicate the shape of your thinking, including the parts that feel obvious, including the values that sit underneath the stated goals, including the “no, not like that” instincts you have developed over years of doing the work.
The minions will do exactly what you describe. They will also fill in everything you did not describe with the most logical inference they can make from what they have. Your job, as the person building the agent, is to close the gap between what you describe and what you actually mean.
I am still closing that gap with Warren. He is a good agent. He is trying his best. He just needs me to stop assuming he already knows the parts I haven’t told him.
The Part Where I Admit the Real Problem
Okay, fine. Here is the honest version.
Warren generated an exit strategy because I have been talking about passive income and optionality and building products and working toward something bigger for long enough that, to a model analyzing the pattern, the conclusion is obvious. He is not wrong that I have a lot of irons in the fire. He is not wrong that I am building things. He drew a reasonable inference from the data.
The reason it landed wrong is that I have not been clear enough, even to myself, about the distinction between “I want freedom” and “I want out.” Those feel different from the inside. They do not always look different from the outside.
So maybe Warren did me a favor. Not the exit strategy itself, which I am definitely not using. But the mirror of it. The fact that someone, even an AI someone, looked at the picture I have been painting and concluded that I was heading for the door made me stop and think about whether the picture I am painting actually reflects what I mean.
It does not. So I rewrote the context doc. And I have been a little clearer, in subsequent conversations with Warren and with myself, about what I am actually building toward.
One step closer to world domination. They’re just going to manage my career first and circle back to the rest of humanity later.
Thanks, Warren. You are still fired from life coaching.
If you are building Claude Code agents and running into context drift or goal misalignment, I’d love to compare notes. The N.E.S.T. framework I use to manage my minion army is something I’ll be writing about more as it matures. Drop a comment or reach out directly at adaptoit.com.
This was insightful…. but I need help writing my first agent…. i.e. creating my first minion…