I recorded a short video this week about Think Big, the Amazon leadership principle that gave my company and this newsletter their name. Many things have happened in AI in the past few months. The principle I talk about has not changed. What it asks of you to do has.
Here is the principle as Amazon writes it:
Thinking small is a self-fulfilling prophecy. Leaders create and communicate a bold direction that inspires results. They think differently and look around corners for ways to serve customers.
When I first wrote about Think Big, my argument was that AI collapses the distance between having a vision and testing it. That statement is still valid. But the last few months have moved the principle onto different ground.
The models now stay on the job
Fable 5, Opus 5, GPT-5.6 and a few others can stay on one job for hours and sometimes days. You notice it as you use them.
Which "job" to hand them has been the harder part, at least for me.
Until recently the pattern was as follows: you gave the AI a specific prompt, got a result, worked with it, went back for the next turn. Some of us had already moved to longer tasks. There is now room to hand over something much closer to a whole job, and this turns out to be harder than it sounds.
Notice that the difficulty has moved. The hard part used to be to get the model to think big enough. Now it's getting ourselves to think big enough.
Two things get in the way
The first challenge is that we are used to being in the loop. We steer, we correct, we watch the outputs take shape. That habit made sense when the output needed watching, and I would not abandon it carelessly. But it was learned under conditions that have since changed.
The second challenge is what this means for trust. If I am not in the loop, I have to let go of a whole piece of work and wait for it to come back. We already do this with people. You give a colleague a job and let them get on with it, checking in here and there. With AI it still feels different, and it makes sense. Possibly it is just that we have had a lot more practice with humans...
I wrote about the related principle in Earn Trust, and about where machine judgment sits on a scale in the five levels of AI autonomy. These become even more relevant now.
What I do notice is that I am handing over bigger and messier work than I was three months ago, and that what I struggle with is usually my own sense of what to give it.
Why this is a Think Big problem
Read the principle again with long-horizon models in mind. "Thinking small is a self-fulfilling prophecy." If the job you hand over is small, you get a small result back, and you conclude the model is good for small jobs. The evidence confirms the assumption because the assumption chose the evidence.
This is the trap we need to escape right now. Not that leaders will overestimate what these models can do, but that they will keep asking for the size of thing they were asking for last year, and never find the ceiling because they never try to approach it.
"Look around corners for ways to serve customers" reads differently too. Cornering takes exploratory hours you did not plan for, competing against delivery of shorter term results. Some of that can now run overnight, which removes the usual excuse.
I have written before about how leadership principles hold up in the age of AI agents. Think Big is the one that has aged best and demands the most. It was always about the size of question you were willing to ask. What collapsed is the cost of asking a bigger one, so any remaining reason for asking small is purely yours.
The uncomfortable version
Executives I work with can usually tell me what their teams are doing with AI. Far fewer can tell me the largest single piece of work they have handed to a model without touching it in the middle.
One caution. I am not arguing for releasing work you cannot evaluate when it returns. That is not pushing ambition but rather abdicating responsibility. Delegate what you can judge later, even when you cannot watch it being made.
There is a second direction here, running work wider rather than longer. I covered that separately in running AI work wide, not just long.
Your action step
Pick one piece of work sitting in your queue that you would normally break into five steps and supervise through each. This might be a full market scan, a competitive teardown, a first pass at a strategy document, or a review of your own pricing.
Hand the whole thing over at once. Write the brief as you would for a capable colleague with whom you have been working for a while: what the finished output looks like, what good means here, what to do when something is ambiguous and how to think about it. Then leave it alone until the model comes back.
When it comes back, pay attention to two things. First, how good the result actually was. Second, and more useful, how often you wanted to interrupt. That shows how much of the constraint is you.
If you hesitated and stepped in before the end, write down what pulled you back. That is a key learning for changing the habit.
If you want to work through where your organization should be handing over whole jobs rather than single tasks, and how to build the review discipline that makes it safe, that is the kind of question I take on in AI strategy advisory engagements and in sessions as an AI keynote speaker and workshop facilitator.