3 ms·
Ask HN: AI writes better code than me. How to keep my identity?
Nowadays, almost everyone codes with AI, right? I am a freelancer, and the entire market is now setting deadlines based on AI-assisted coding speeds. If you don't use AI, you simply cannot meet the deadlines. Because of this, I completely transitioned to AI-driven coding at the beginning of this year (starting around GPT-5.3).
It's not that AI coding isn't fun. It is. But in the beginning, there were still many things AI couldn't do well. The community focus was on "Harness Engineering," attaching MCPs, or deep Prompt Engineering. I actually built and used quite a few custom AI tools during that phase.
In the past, you had to provide not just your intent and direction, but also detailed, step-by-step instructions to reach that goal. Lately, though, even if you skip the detailed instructions, both Claude and GPT-5.6 handle it perfectly. You don't even have to agonize over which model to use. "Graph engineering" is currently trending, but it seems the models have already learned those patterns natively—if you establish a high-level plan with them, they automatically embed the execution policies into it. Whenever I try to learn and apply a new AI engineering technique, the next model update already does it out of the box, making the effort feel pointless.
The real problem is that AI is now too good. It applies ADT patterns, immutability, Option/Maybe, Result patterns, functional programming patterns, monadic chaining, and architectural structures like MVC, MVVM, MVI, and Hexagonal architectures far better than I can. It's heavily demotivating.
All those years I spent memorizing GoF patterns, slowly applying them to my work, and tweaking them into my own style—it all feels meaningless now.
Honestly, the vast majority of freelance work revolves around simple CRUD apps. Sure, there is domain knowledge to memorize, but when it comes to structuring CRUD, using functional techniques for error and value handling, or writing boundary tests, AI is just far more meticulous and capable than I am.
It even feels like AI is better at domain modeling. I try to hypnotize myself into believing I'm still necessary, but realistically, I don't work in Big Tech. I'm a programmer who mostly assembles business logic on top of existing frameworks, and the truth is, AI does my job better. And faster. I don't think I can deny it anymore.
Top-tier developers might still outperform AI, but for the vast majority of us, including myself, that is no longer the reality. To write the kind of code AI generates, I used to have to keep documentation open on a second monitor, think deeply, and constantly Google better approaches. AI just generates it instantly, making my past efforts feel obsolete.
When AI writes better code than I do, how am I supposed to maintain my motivation and identity? I would especially love to hear from other freelance developers. With market deadlines getting tighter and tighter because of AI, how are you handling this transition?
Again, it's not that AI "vibe coding" isn't fun. It is. But entirely separate from that, the stark realization that I am losing my competitive edge is making me increasingly depressed.
- minimaxir 29d agoWriting code is only one part of software engineering. Making good engineering decisions is the other, and LLMs aren't optimal at that...yet.
- diavelguru 29d agoHow long have you been working in this field? I ask as if you have numerous years you already know your worth and value irrespective of AI or any other system. If you are still questioning this it indicates to me that you are still in the process of finding yourself and are in the same boat many other folks are in - confronting the existencial crisis of identity as it pertains to work. I suggest do what is necessary to stay employed then on your own time write code by hand for personal or side projects. Be sure to not use AI and learn deployment, devops and infrastructure as much as is needed to be knowledgeable and able to speak to the technology. Then add that to your resume. Keep grinding.
- jdw64 29d agoI'm currently a programmer with 8 years of experience(I've been coding for longer than that, but I started earning money from programming around then.). Some might say that's still short, but honestly, I'm not sure. I became a programmer during the COVID-19 hiring boom, and somehow I've managed to survive. But I'm not really sure what I should be doing next.
- oliculipolicula 27d agoWhat you should be doing next is to pivot your programming skills plus interest in identifying "categorical errors" to help me look for elephants in the room housing all the different "categorical imperatives" :) as a hobby even. Because my not very adult hunch is that co-existence with one another and nature depends on paying attention to these animals >Our world relies on outsourcing the production of manufactured and consumer goods to developing nations. But separate from that structural reality, vested interests refusing to acknowledge their faults is an entirely different issue. [faults= giant problems with deeply held but actionable beliefs] There seems to be already two different instances of CI in that quote... To extend/detail on the second one, make it relevant to this website--- What if AI helps to make it easy for "everyone" to build things people want? That doesn't bode well for the climate. If I were one of the local gods I'd rationalise (ie sidestep CI) that by saying that hardware is so hard that maybe only 1% of the population in developed countries would be able attain this kind of (post-)industrial upper middle class [0] [I tend to see categorical imperatives as a the general way to obtain constraints on action local in spacetime by extrapolating to extreme hypotheticals, more physicist than "edge cases". This is one understated belief that (I feel) silicon valley thought leaders tend to subconsciously share with economists, neuroscientists[1], psychologists, techbros, journalists, rationalists---- but not Wall Street---- the sanity of linear utility. 2] https://youtu.be/e5O1Y-5phEU https://youtu.be/e5O1Y-5phEU [1] if this come across as a collective but ad hominem, I'm hoping it still looks like a reasonable one ;) anyways i probably damaged my credibility by submitting a neuroscience scam that time (not related to, but reminiscent of Parasma) [2] I'd even say that you can't renormalise linear utility, just like one "would not be able to renormalise some Hamiltonians" so the reflexive rationalisation [0] fails, if you care to go there