The latest AI drama is not a founder launching a new app or a billionaire changing his profile photo. It is a researcher walking away from one of the most important labs in the race and saying, essentially: the brakes are not keeping up with the engine.
Jacob Coxon, who said he spent the last three years doing pre-training research at OpenAI and Anthropic, announced that he was leaving Anthropic over concerns about how quickly the industry is pushing toward more capable, potentially self-improving systems. The Associated Press reported that he believes the labs are competing to build the most advanced models while safety work is not keeping pace.
That is a very large claim. It is also a very familiar business problem: the people closest to the work are telling leadership that the system is moving faster than the controls, and everyone with a quarterly target is pretending that is a later conversation.
If your growth plan needs everybody to stay calm while the people building the thing are quietly panicking, it is not a growth plan. It is a group chat with a deadline.
Slow is not the point. Control is.
I am not interested in turning this into an AI panic column. The current systems are useful. They help people write, research, code and run businesses. The point is simpler: when the downside gets enormous, “move fast and fix it later” stops being a personality trait and starts being a risk policy.
Reporting from TechCrunch and WIRED describes Coxon's concern around recursive self-improvement — systems helping build more capable systems — and the lack of confidence that the industry knows how to align a future version before it is too late.
Maybe his timeline is wrong. Maybe the risk is lower. Maybe the industry figures it out. None of those possibilities make “we will know after we ship it” sound like a serious operating model.
The business lesson is painfully good
Every woman who has grown a company has faced a smaller version of this. A team wants to take on more jobs before the crew is ready. A founder hires faster than the systems can train. A sales target gets set before anyone checks whether delivery can keep the promise. Everybody calls the speed exciting because nobody wants to be the person who says the machine needs a better brake.
Here is the tea I would steal from this mess:
- Set the red lines before the launch. Decide what you will not trade away for speed while everyone is still feeling optimistic.
- Give dissent a promotion path. The person who sees the failure mode first should not have to resign to be heard.
- Track incidents, not just wins. Revenue tells you what happened. Near-misses tell you what is coming.
- Own the downside. If nobody can name who stops the rollout, the rollout is already in charge.
The internet loves a visionary who says she will figure it out in public. Fair. That is how a lot of women build something from nothing. But there is a difference between taking a personal risk and asking the whole world to absorb the consequences of your unfinished controls.
You do not have to be anti-AI to demand a business plan with a kill switch.
Coxon's resignation is a warning, not a verdict. I do not know whether AI is moving toward a catastrophe, a breakthrough or a messy combination of both. What I do know is that “faster” is not a strategy by itself — not in a construction company, not in a cleaning business, not in software, and definitely not with technology that may eventually help rewrite its own playbook.
Build boldly. Ask better questions. And if the smartest people in the room are asking where the brakes are, maybe do not answer them with another launch date.