A Working Thesis

The next important opportunities in technology will come from combining deep software knowledge, AI, and hardware. That is the bet I am making.

I have been lucky to work on both sides of a transition. I wrote code by hand for years. Now AI writes a large part of it. That shift has opened doors that did not exist before. Problems that once felt too hard or too slow to explore are now much easier to get into.

I do not just mean writing code faster. I mean going deeper. Understanding why a system works the way it does, seeing where it can fail, and building the thinking you need to reason about the whole thing.

There is an advantage in having learned before AI got this good. We typed the code ourselves, struggled through hard parts, debugged by hand, and saw what happened beneath the abstractions. That built stronger mental models. Writing code by hand, especially early on, still matters for building judgment.

AI does not yet think the way humans do. There is a kind of thinking that comes from staying with a hard problem for days or weeks. Over time you notice patterns and hidden connections that were not obvious at first. You build intuition through repeated thought, failed attempts, and close attention.

AI can help with this. But it cannot yet replace the judgment you get from living inside a problem until you understand it from the inside. People who have never spent that long on a single problem underestimate how powerful this kind of reasoning is. Use AI as a tool, but do not give up your own ability to think.

Meanwhile code is becoming abundant. When almost anyone can generate working code, producing code stops being scarce. And what stops being scarce stops being valuable on its own.

Software will not matter less. But the software industry in its current form will shrink. Companies built around large teams writing routine application code will get smaller as AI drops the cost of implementation. At the same time new kinds of software will appear. Value will move toward systems that combine software with AI, hardware, infrastructure, and science. How software gets built, and which skills command the most value, will change.

I expect this to become obvious over the next four or five years. The hard part is that we are in the middle of it. The old world is already changing, but the new one has not fully arrived.

Most research and investment today still goes into making models more capable. I believe this will plateau in practical terms. Not because models will stop improving, but because we will reach a point where a reasonably capable model is good enough for most tasks.

Once that happens, the focus will shift to building systems that combine software, AI, and hardware. By hardware I do not just mean robots. I mean rockets, space systems, edge devices, scientific instruments, and every other place where software meets the physical world.

The new baseline for a strong engineer will be someone who understands software deeply enough to know where systems fail, who knows distributed systems, who understands AI beyond surface tools, and who can apply all of it to problems in the physical world. Understanding AI does not just mean building agents. Agents are one small part of the field.

Much of what we learned building software will still matter when we work with hardware. Reliability, latency, networking, failure handling, and distributed execution do not go away. They matter more, because the system no longer lives only inside software. It touches real machines, physical constraints, and mistakes you cannot fix with a retry.

I believe companies will form around exactly these problems. Some may already be moving this way.

This is where I stand. I know how to build software, scale systems, and take a product from idea to working thing. The next step is to go deeper into distributed systems, then machine learning, then hardware.

The exact path is still blurry. I expect to change parts of it as things get clearer. I am not pretending I know the destination. But I believe the direction is right.