AI is leverage, and leverage works in both directions. The same tools that let you multiply your productivity can just as easily multiply your damage, because leverage doesn’t care what you point it at. It amplifies whatever you bring.
You can feel this most clearly in coding. With a modern agent like Fable, you can generate an enormous amount of code in a single sitting, more than you could have written by hand in a month. But all of that code is now an artifact you have to steer after the fact, and the more you generate, the larger the thing you have to understand, review, and guide. The output grows faster than your attention does, and you’re left holding something much bigger than what you actually typed.
This is decompression. You write a small, dense source, a prompt, an intention, a plan, and the agent expands it into a massive delta of code. A few sentences become a thousand lines. A small ambiguity in the source becomes a large divergence in the output. The compression ratio is huge, which is exactly what makes it powerful and exactly what makes it dangerous, because a tiny error or a fuzzy idea at the source doesn’t stay tiny. It expands right along with everything else.
So the question becomes, what are you compressing? If the source is muddy, the expansion is muddy at scale. If the source is clear, the expansion carries that clarity into every line. The leverage lands on whatever you put in.
This puts more weight than ever on the things that were always quietly important. Deep understanding, so you actually know what you’re asking for. Clarity of communication, both to the agent and to the stakeholders around you, because a vague instruction decompresses into a vague system. Depth of planning, so that the decisions you make trace cleanly into the actions the agent takes. And a longer time horizon, the ability to connect the dots across weeks and months and many expansions, because each delta you generate becomes the source for the next one.
What this gives engineers is the chance to work at a higher level of abstraction than before. You spend less time on syntax and more time on intent. But it would be a mistake to read that and conclude engineering is solved. Writing code may be close to solved. Engineering is not. Engineering is the act of deciding what to build and why, of holding the whole system in your head, of choosing well under constraints you can’t fully see. The leverage doesn’t remove that work. It raises it.
There’s a defeatist way to look at all this, which is to say that if the machine can write the code, then humans aren’t needed anymore. I understand the instinct, but I think it gets the picture backwards. A longer lever doesn’t make the person holding it smaller. It lets them move something heavier. We can reach for higher abstractions and harder problems than we ever could before, the kind of problems that used to be out of range simply because we ran out of hands.
And we bring something to the lever that the lever doesn’t have on its own. We bring the motivation, the reason any of this is worth doing. We bring the empathy to know which problems actually matter to people. And we bring the willingness to do the hard, unglamorous work of pushing on a real problem in the real world until it gives.
Do you believe we will ever solve all of the problems? I don’t even consider it a possibility. There will always be a next problem, a harder one, one we couldn’t see until we cleared the one in front of it. That isn’t a sad thought to me. It’s the opposite. It means the work never runs out, and the reach keeps extending.
That’s why the optimistic take on this AI-accelerated world isn’t naive, it’s earned. There is a real rush of excitement in suddenly being able to reach for the hardest problems we have, the ones that have always felt too big. How do we take care of our elderly. How do we give people more time with the people they love. These are the problems worth all this leverage.
We bring our time, our attention, and our care. The machine brings the reach. What a time to be alive.