Kyle HughesDec 31

The two biggest things I got wrong in 2025:

1. AI would not be able to write 100% of my code by end-of-year.

The space is moving faster than I expect even though I am credulous. It is also not moving linearly—AI could not write 70% of my code in September so I did not see how it would arrive at 100% in 3 months. From my perspective it happened overnight as emergent behavior.

https://mister.computer/@kyle/114246913696659286

Kyle Hughes (@[email protected])

If you can get beyond the undeniable threat of these statements—coding will be obsolete by the end of the year, AI will be writing 90% of all code in 5 months, 100% in a year—it’s intriguing that these timelines are extremely short with verifiable metrics. The richest and most influential people in the world are guaranteeing that they will have built God by September and no one talks about the implications or takes them seriously, even their investors. That’s strange. https://techhub.social/@Techmeme/114246319851894848

Mister Computer
Show threadKyle Hughes

2. AI writing 100% of my code means there is no value for me to add.

I now spend all of my engineering time making novel, high-value architecture decisions—more in a day than I used to in weeks. It is turbo software-making. I have a sensation of working at the top of my license that I also would have attested to before, but now I see what I am capable of when I don't have to also write the code. More energy and capacity for decisions means fewer compromises across the entire stack and product.

Dec 31 at 5:54pmWeb
Show threadKyle HughesDec 31

I am most confident that in 2026 the cost of all of the software we are making will plummet. I have no idea what the implications will be. It could be cool. We should be able to get past rendering JSON in list views, as an industry.

I make $100/hr. To add a floating user selector to a chat text field, with all of the trappings and polish and testing, probably takes me two days. $1,600. I did it with Claude Code in 3 hours. $200/mo overhead + $300 in my time. I also did other work in parallel.

Show threadKyle HughesDec 31
At home, I am modernizing Information Superhighway—for the first time in two years—while also adding features to Rank Things and building another new indie app, all at the same time. Literally the same time—three agents running in parallel and my machine going nuts with builds. I have so much downtime that I have to start new projects and increase scope of others to fill my time. It is abundance, and these are only the beginnings of the second-plus-order effects. They’re falling out of my head.
Show threadNick SloanDec 31

@kyle But you’re selling your clients code that they can’t own, right? My understanding is that the output of LLMs is not copyrightable.

Maybe that doesn’t matter for your projects, but I’m really curious to hear how businesses will reckon with that.

Show threadNick SloanDec 31
@kyle Or perhaps Congress will say that actually LLM output is copyrightable and the creators of the model own its output.
Show threadKyle HughesDec 31

@nick I am a salaried full-time employee, the math was for convenience. As a company we are comfortable that the products we are building have substantial protectable elements across copyright and elsewhere: human requirements, human review, human modifications, human architectural decisions. Even if some constituent parts of the codebase are not protectable.

To the other point, Anthropic already has commercial terms to protect against that.

Show threadNick SloanDec 31
@kyle Thanks for the reply and taking the question seriously. I’m still finding that this stuff pushes up against my risk tolerance, and I’m still on the fence about the ethics, but I appreciate very much sober conversations with people who are further ahead on this than me.
Show threadKyle HughesJan 1
@nick that’s what I’m here for 🫡
Show threadjoe kimberlinJan 1
@kyle the craziest part (and most valuable to me?) is how it saves me from essentially starting from scratch in other areas i have zero knowledge of. i’ve wanted to make an android version of my main app for 5 years and the cost has always been too high. until now, where claude has taken my iOS code and made a (hopefully??) shippable android app with minimal supervision
Show threadKyle HughesJan 1
@Joekw This feels like where a large part of the guaranteed acceleration will come from: you only need an example of something done well once. Whether that’s an implementation of an abstraction, or an explanation of a tricky concept from a blog post, or an app, or a document that describes how to grow your architecture… Once you have captured that essence you can continuously reuse and reapply it without having to remember to. And then that keeps snowballing—only do anything once. Forever?
Show threadHalDec 31
@kyle sounds like you are adding a lot of value!
Show threadKyle HughesDec 31

@hal I definitely am. I made sacrifices and put a lot of effort into getting myself in a position to maximize my value if this moment were to come to pass and it is working. And my effort is the tide that is rising all boats around me.

But more questions are unfolding. What to recommend to students? How to help others work through their ego to get here? What happens to the livelihood of people who find this mode of working distasteful? Can this thing I am still good at also be commodified?