What Will We Do When AI Writes All the Software?
AI has gained incredible skill at coding — I think we all agree on that. And extrapolating this trend, it seems clear that we'll use virtual AI assistants much like real people. As capitalism dictates, labor always gets cheaper relative to its output (i.e., productivity increases).
At this point, even though a few things are being overlooked — as I mentioned in the previous post in this series — we'll ignore them for now too. To avoid repeating myself, I won't revisit those points; if you're curious, you can check out the previous post [s1].
Today I want to talk about something different. I want to talk about the death of knowledge-shuffling.
The End of Knowledge
I want you to imagine a scenario. AI has truly gained persistent memory — it no longer acts like it has Alzheimer's every session. We've integrated it into real software systems, given it tasks and personas. AI has taken over all white-collar jobs. Every job in the digital world that involves shuffling knowledge has given itself over to the cheaper alternative: AI. Just like "all" heavy labor was handed off to machines... except it wasn't. Machines didn't take all physical labor, because in some jobs humans are still cheaper or more adaptive. But in software, both of those defenses collapse: AI's cost is approaching zero, its capability keeps growing, and in the digital world there's no adaptation problem. This time, the conditions really could be different.
Let me clarify what I mean by "shuffling knowledge": writing code, preparing reports, analyzing data, drafting contracts, designing presentations, business correspondence, etc. What I really mean is "processing information in the digital world" (I just avoided the word digital because there are also information-processing jobs that haven't been digitized yet but could be). If you define your job as sitting at a computer and thinking, you'll probably need to interact with AI.
We're not yet at a point where all jobs have been handed over to AI — I'm aware. But the direction is clear. Less and less of value is built on "manipulating knowledge" with each passing day.
So what does this shift mean, and what should we do?
The Beginning of Wisdom
I want you to think about what it means for AI to handle all "white-collar" work. Not to mention the blue-collar jobs that will follow with robotics.
Shuffling knowledge will become completely worthless. I want to repeat this sentence, because you can unlock this entire blog post just by understanding it.
Shuffling knowledge will become completely worthless.
So what will be valuable? Let's list some options — I'd like you to read these with a wide-angle lens:
- Knowing which knowledge is worth shuffling
- Knowing what to build becomes more important than knowing how to build.
- Shuffling knowledge the fastest
- When everyone has the same competitive tool, speed and volume become the decisive factors.
- Holding the most knowledge that can be shuffled
- How can you compete with Google? Knowing things about your customers that even they don't know about themselves could yield interesting results in a world where the cost of software drops to zero.
- Obtaining truly original knowledge that nobody even has an imitation of
- You go so deep in a certain field that you gain a competitively advantageous insight. In an era when everyone has AI tools that can code with cutting-edge technology, developing the "cutting edge" itself might make more sense than focusing on AI.
- Bringing shuffled knowledge to life through real-world relationships
- AI can be smart, but is it effective? Nerd vs. social. Technical success doesn't equal product success.
I want to touch on each of these points from the perspective of "how does this affect us — the people who love building things." A single reality can be examined from different perspectives. Let's do that:
The Value of Knowledge
The question of what should and shouldn't be built might seem trivial, but it's actually crucial. It's a problem that exists independently of AI, and many companies have gone under because they couldn't solve it properly. We have no measurement method for what's valuable to build. No theory either. If we had one, it would probably lose its validity anyway (likely due to the efficient market hypothesis). Humans are fascinating creatures, and the reality they live in changes over time — states, commodities, companies, players all shift. What's worth building will always be a question intrinsic to people and specific to context. Even a general AI can't understand the context humans are in without interacting with them. (Part of what makes this point so hard is that even if you're a smart person, you have to look up from your computer to understand what's worth building.)
So anon, touch some grass. And learn what's worthy to build.
Speed, I Am Speed
I'll make another reference to the efficient market hypothesis. When everyone has the same tool, competition has to burst out from somewhere. This probably isn't a perfect assumption (that everyone has the same tool and the only differentiator is the builder's speed), but I don't think it's a terrible one either. At the end of the day, speed will matter. We all know this. Like the previous point, what's interesting here is that it's also a general principle. Independent of AI, how fast things are built is critically important in the market — in some cases, it's a life-or-death matter for a company.
After you see that something is worth producing, the task isn't just to produce it — it's to produce it immediately. When something goes wrong, you need to take action immediately. When something goes right, you need to apply leverage immediately. As tools get more powerful, reflexes become even more important.
Quantity Over Quality
This might feel counterintuitive (it's a point that even disgusts a perfectionist like me), but from one perspective of reality, quantity trumps quality. In my view, this is one of the reasons large companies like Google manage to succeed at things. Big companies do almost nothing perfectly (apart from their core service / differentiating factor), yet they still dominate. How?
YouTube isn't perfect — far from it. Tons of design improvements could be made. Tons of UX improvements could be made. Do you think Google is stupid? Why don't they do it? Because perfectionism is the greatest rival of being good. Because large companies understand that being good across many areas is far more important than being perfect in one. That's why Apple has a ton of hardware products. That's why YouTube is trying to become "everything TV." That's why Microsoft Office is determined to also be your browser (and Google competes in kind).
But here's what's really interesting — there's a byproduct of this "broad but not perfect" strategy: data. When Google Maps launched, it wasn't perfect, but millions of people used it. As they used it, it collected traffic data. Traffic data made it indispensable. YouTube's content quality still isn't perfect, but billions of views gave Google a matchless pool of viewer behavior data. Google knows things about its customers that they don't even know about themselves. Because instead of trying to be perfect, Google tried to be broad. The broader it got, the more data it collected. The more data it collected, the better it learned what mattered. The more it learned, the better it produced.
This is a flywheel: produce a lot -> collect a lot of data -> know better what to produce -> produce more. When a perfectionist stands still, the flywheel doesn't turn. A flywheel that doesn't turn produces no data. A system that produces no data stays blind. A blind system can't know what's valuable.
From a finance perspective, you can look at this as "expanding the playing field." When a company dominates one area, it tries to expand into others. When doing so, it doesn't try to be perfect — it just tries to be good enough to capture market share. When it's good enough, the time to dominate may come (Google -> Maps, Microsoft -> Azure). But that domination would never have come without entering the field, collecting data, and spinning the flywheel.
The takeaway from this point (assisting the previous ones too):
Once you've learned what's valuable and how to do it fast, the next thing to do is repeat. Repeat broadly, without trying to be perfect. Every repetition gives you new information; every piece of information makes the next repetition better. Until you succeed. If you stay alive long enough, you will succeed.
Quality Over Quantity
The more poetic point — the one nearly all of us pray to be true. Even if AI takes everything from us, it can't take creativity. The differentiating factor, the next-generation breakthrough technology, the knowledge nobody else has... it can't know these things. The value of building these factors will increase exponentially. If the previous three points advised you to go broad, this one advises you to go deep. Truly love what you do, and try things that nobody could even conceive of, that they'd be too lazy to attempt. Work hard, strive, and contribute new things to humanity.
I don't think the value of this will decrease no matter how much AI advances — in fact, AI could even serve as a lever. A lever that amplifies the importance of new technologies. When a database runs better, its contribution to humanity becomes far greater. When a video compression algorithm runs better, its impact becomes exponentially higher. More examples could be given, but the core idea is the same.
My belief that human effort will be completely discarded is zero — negative, if possible. If a person is genuinely adding to themselves out of curiosity, and can then share that contribution with humanity by finding new ideas, there will absolutely, without question, be a reward for this in every era that humanity exists. Societies that fail to reward this are doomed to perish. So don't shy away from learning just because AI exists. Don't see your questions and your work as foolish. Don't say "AI can already do this." Because when you reach the ability to do what it can't, you probably can't even imagine the leverage you'll have.
Everything Is Just Marketing
This might be the point you'll push back on the most. For technical people like me especially, the word "marketing" is almost an insult. "I'm an engineer, not a marketer." I get it. I used to feel the same way.
But when you truly enter the market, you see that questions like what you can do, how you do it, and why you do it all get crushed by a single question:
What can you sell?
At the end of the day, how good your work is doesn't matter — what matters is who knows about it, who'll pay for it, how long it stays relevant, how the product evolves going forward, and so on.
As engineers, we love looking at everything purely through a technical lens — if the solution to a problem is good, everything is done. This flawed perspective stems from our generally flawed outlook on life. Problems don't exist to be solved; they exist to touch human lives. You can solve the hardest problems in the world, but if they don't touch anyone's life, nobody cares. No matter how elegant your solution is, if it doesn't touch anyone's life, it means nothing.
That's why, when you finish a technical task, you've only completed half the road (sometimes less than half). The rest is telling your story. Selling it. And usually, this is something that should be thought about before you even start the technical work. If there's no way to sell what you're going to build (worst case: personal use), there's no reason to build it either (at which point you can circle back to point one).
At the end of the day, AI can write your code. But it can't tell your story. It can't sell it.
And yes, everything is just marketing. Because a tree that nobody hears fall makes no sound. In any reality.
How Does One Become Wise?
Wisdom is not like knowledge. You can't download it. You can't write a prompt and become wise. This is precisely what makes wisdom so valuable in a world where AI can shuffle all the knowledge.
So where does wisdom come from? Maybe you've already figured it out: experience.
If you look at the five points above again, you'll see that they all say the same truth from different angles: look up from your computer, engage with reality, do something, learn from what you did, do it again. To understand what's valuable, you need to talk to people. To be fast, you need to try. To collect data, you need to produce. For deep knowledge, you need to genuinely wonder and ask. To sell, you need to understand the person in front of you.
None of it can be obtained by asking AI. All of it is obtained by doing, by trying, by engaging with reality. You must fail. That's the only way to become wise. Being wise means daring to try things. Being aware that you're inexperienced.
Wisdom is a cycle: do -> fail -> learn -> do again. Every turn of this cycle makes you a little wiser. AI can accelerate this cycle (you build faster, test sooner, access knowledge quickly, get feedback faster), but it can't enter the cycle for you. Only you can turn the wheel.
That's why AI can write everything, but it can't write you.
You must write yourself. You might do a terrible job at first — that's okay. What you learn from terrible will always be more valuable than what you learn from imagining perfect.
Knowledge can be downloaded. Wisdom is earned. AI will drive the first to zero. Nothing will drive the second to zero.
NOTE
What should programmers do:
- Care about learning — depth will always matter
- Build speed — experiment fast, fail fast
- Keep your horizons wide — don't get lost in a single domain while going deep
- Stay social, look at the world -> Be a storyteller
- Shift from an engineer mindset to a builder mindset -> Don't dwell in problems / reasons, deal with the reactions and people too