Hello!
It's Thursday, 21st May 2026 and you are reading Bold Efforts!
The most common thing I hear from operators right now is that AI adoption is slower than expected. What I find interesting is that nobody blames the models. The tools are capable. Something else is in the way.
In 1847, a Hungarian physician named Ignaz Semmelweis noticed something that should have been obvious. The maternity ward at Vienna General Hospital staffed by doctors had a mortality rate nearly five times higher than the ward staffed by midwives. The doctors came directly from performing autopsies; the midwives did not. He introduced mandatory handwashing with chlorinated lime solution. Mortality dropped almost immediately. The evidence was unambiguous.
The medical establishment ignored it for decades.
Accepting Semmelweis meant accepting that doctors had been causing the deaths they were supposed to prevent. That reckoning, moral and professional, was what the establishment would not absorb. The practice changed only after his death, and slowly. The knowledge existed long before the behavior did.
Machiavelli identified the underlying logic five centuries ago. The innovator, he wrote, faces fierce resistance from those who benefited from the old arrangement, and only tepid support from those who stand to gain from the new one. The asymmetry is structural: losing feels immediate, gaining feels hypothetical. Those who lose fight hard. Those who would win wait to see if it holds. Change stalls in the middle.
This is the condition AI is walking into.
Every organization I have spoken to over the last year has a version of the same situation. The tools are deployed. A few people use them with real discipline. The rest continue as before. When you ask why, the answers are almost never about capability. The models are good enough. The use cases are clear. What is missing is the shared agreement to change how work gets done.
The most common failure mode: the tool surfaces a recommendation, and the team files it away. The reason is usually something like this: acting on it would require us to change who owns this decision, and we have not had that conversation. Or: the answer is right, but following it would make someone's role redundant, and no one has agreed to that yet. The AI found the answer. The organization could not get to yes.
There is a pattern in the AI deployments that actually work. What they share is clearer authority. Someone with enough standing decided: this is how we work now, and relitigating it is not on the table. The willingness to decide produced the outcome. The AI made the decision harder to ignore, but someone still had to stop ignoring it.
This is the adoption paradox. The organizations most likely to benefit from AI tend to have the most complex coordination challenges. Complex coordination challenges are precisely what makes changing behavior hardest. The tool works best where the ground is already prepared. It works least where the need is greatest.
Buying the tools is the cheaper part. The organizational work that has to come first is expensive: who gives up decision authority, who absorbs the accountability shift, who the system is actually optimizing for. These get answered in rooms where someone has enough authority to end the debate, and chooses to use it.
Semmelweis had the data but the hospital had the culture. And the culture won for thirty years.
Intelligence, in organizations, was always cheaper than it looked. We consistently underinvested in the harder work: the conversations, the trade-offs, and the moments when someone had the authority and willingness to stop deliberating.
AI makes intelligence cheaper still. The constraint it cannot touch was always there. Thank you for reading.
Best,
Kartik
I write Bold Efforts every week to think clearly about where work and life are actually headed, not where headlines say they are. If you want these essays in your inbox, you can subscribe here.