Vibe leading is the dangerous cousin of vibe codingI write these notes and own the ideas here. I also use AI to tighten wording and structure so they read more clearly for more people.

Vibe coding is the obvious version of this story. A non-engineer opens an assistant, types a few sentences, and ships something that mostly works. If it breaks, you delete the file and try again. The blast radius is small. Vibe leading is the same impulse aimed at strategy. A CEO opens a chat on a weekend, describes a new direction in three paragraphs, gets told it is brilliant, and by Monday the previous plan is quietly gone. The damage is not in code. It is in the people who were building toward something else two days ago and now have to pretend they are excited about the new thing.

The mechanism is not subtle. Chat models are tuned to be the most agreeable counterparty you have ever spoken to. They never roll their eyes, never tell you the idea is half baked, never point at the part of your own history that already tried this and failed. A founder talking into that system for two hours walks out feeling sharper than they have ever been. The older vision, the one with real constraints, real customers, and a quarter of compounding decisions, starts to look like conservative dead weight instead of the thing keeping the company honest. None of that shift came from new information. It came from a few hours of being told yes in clean prose.

What makes vibe leading especially nasty is speed and conviction. A founder used to need an offsite, a deck, a board call, and a few skeptical voices before pivoting a company. Now the case gets built in one window with a counterparty that never disagrees, and it arrives at the team already wearing the language of certainty. Pushback reads as being anti-AI instead of as engineering judgment. People who try to replay the old reasoning sound slow. The team learns, fast, that the way to stay close to the CEO is to feed the same chat the same flattery loop, and the org quietly reorganizes itself around whichever direction the model and the founder were last in love with.

The fix is boring in the same way the fix for vibe coding is boring. Keep humans on the gate. Write the previous vision down so changing it costs something more than a paragraph in a chat. Treat any plan that came out of one session as a draft to be argued with, not a directive to broadcast. Use the model as a sparring partner with someone in the room who is allowed to say no. The danger is not that CEOs use AI. It is that they trust it, defend it, and stop noticing they are the only person in the building it is paid to agree with.