One thing that strikes me about what is going on: we have been promised this renaissance in software, yet, it been 2-3y of accelerated agentic/LLM development and I haven't see anything great coming out of the LLM developed software world that is worth looking at. Let's take open source as an example: No new 3d/2d CAD packages that are more capable than Blender/FreeCad/KiCad, nor Blender/FreeCad all of the sudden became much more capable than they were just few years ago. I also want to point out,that I haven't seen any new LLMed software in Electronic/ElectroDynamic simulation, something I would LLM to excel at, beyond what already existed before. No new software that is equal to Adobe Creative Suite (that Vibed coded replacement is trash IMO), No new Operating Systems that are taking over. No new utilities that are super useful in day work. We are using the same software, we've used before LLM arrived.

Reverse Engineering became much easier, it is plus. Finding vulnerabilities also became much easier, another plus. Models are getting better. But where is the new cool software ?

Edit: I will go even beyond OSS. M$ & Goog, who are DEEPLY Vested into AI, have access to the best possible frontier models and large groups of highly proficient engineers. Yet, Windows and Android are as laggy & buggy, as they were 4-5 years ago.

Hmm, that may well be true but I don't model with any of those packages. I model with `build123d` or `openscad` using Claude Code to do the heavy lifting. I describe what I want, measure and make up ideas and learn a little terminology (fillet, chamfer) and then the machine builds me the concept that I then print and deploy into action[0].

My life has been improved immeasurably (our home is filled with attachments and affordances built on-demand by Claude), and no new software was created that anyone else would see. It is entirely possible that these tools fail to create a Photoshop/Blender replacement and yet replace Photoshop/Blender.

My wife started out building in Blender[1] but these days even she uses Claude Code and mostly ideates. The nature of LLMs might not be "they can build you something better than Photoshop in a day" but that "where we're going we don't need roads".

0: https://wiki.roshangeorge.dev/w/Blog/2026-01-11/Modeling_Wit...

1: https://wiki.roshangeorge.dev/w/Blog/2026-01-15/Modeling_Wit...

One of the things I've started noticing at work is that for the last few months I was banging away at having multiple streams at a time. But over time, I've reached the point where there is some human blocker on the stream, be it QA to test it, other humans to have attention to pick it up for the next step, deployment of some sort, or in some happy cases, it made it through to deployment and the problem is now solved. In the past couple of weeks my multistreaming is toning down and more often I'm back to one stream of work. Often a much higher level stream than I would have been working at a couple of years ago, and I'm moving faster in a number of ways. Some of that "faster" has been burned up in things like changing environments that I previously just wouldn't have done, though.

Open source has had a QA problem pretty much its entire life... not in the sense that it stinks necessarily, but in that getting it at all is a challenge and was in many projects probably already the blocker to further progress.

One wonders if what is happening here is just that even in the software space, where the fluidity and flexibility is maximized as we work with nearly entirely digital constructs, that once we burn through the dry tinder of "all the stuff we know we needed but didn't have the coding time to get to" and the frontier of "work to do" advance on all fronts to "some sort of human or real-world blocker", it may just turn out that the steady state improvement in productivity after accounting for Amdahl's Law, even when code is "nearly free", is only a maybe 2x or 3x improvement after all. If this is true, goodness help the financial bubble which is assuming that something like 5x or 10x indefinitely is a lower bound, if even in the best possible domain for AI that's all we really get.

> buggy, as they were 4-5 years ago

I wish!

The quality of commercial software has really gone downhill the past decade. Some of it may be AI, but it has likely more to do with the fact that fewer and fewer want to pay for QA.

On the endless framework treadmill, money is better spent going forward and changing things.

I thinks what's more happening is bespoke software made by people for themselves mostly, often with a user count of 1. That's quite a huge change, for users of that software. Of course such things are not replacing existing apps.

I think this comment is premature. It hasn't really been 2-3 years accelerated LLM development. Even 6 months ago, LLM development was much less refined than it is now. There's been a few years of figuring out how to use these tools (which are constantly changing) and maybe a year of using them effectively.

You point all these new software but some of them are literally only a few months to a few weeks old and complaining that they aren't up to standard yet. Even if LLMs didn't get any better than they are right now, these projects have barely even begun, and there is plenty of time and resources to build them up.

I've written a bunch of cool new software with LLMs in my spare time -- are you sure you even have a way to notice all the new cool software that exists?

We have seen a huge proliferation of vibed audio plugins. Most of them are very bad or replicate things DAWs can already do. And vibed video games. If anything it has become harder for many audiences to find good software because algorithmic curation is driven by gameable engagement metrics not quality.

Yes! It's just repeatedly asserted that vibecoding is taking over, going to replace us all. I have yet to see evidence of it (no, marketing stunts aren't evidence).

I personally think the agents and the harnesses themselves are pretty cool software. Claude code, openclaw, hermes, and now the consumer facing ones. I have been doing all kinds of cool things with them. If you gave somebody the claude desktop app with blender mcp 10 years ago their minds would be absolutely blown.

A lot of these harnesses and mcps themselves are built with llms.

> But where is the new cool software ?

No matter what you think about it, it's impressive what the folk at Omarchy[1] have been doing with LLMs.

1: https://omarchy.org/

This is the great paradox that many people have observed: why are we still using frustrating, buggy, hard-to-use, expensive software all day?

We're pretty much still building the tools to build the software renaissance.

It's here already, but not at a global scale.

You can pick any application you pay for, but only use 10% of the features of, and create a bespoke version that only has that 10% in it. It'll take you a day to get an MVP and maybe a week or two for it to be "good enough".

> No new utilities that are super useful in day work. thats the renaissance for you - you no longer need the utilities, just ask the agent to do that, whatever it might be.

Do you count ArtCraft?

> Reverse Engineering became much easier, it is plus. Finding vulnerabilities also became much easier, another plus. Models are getting better.

I think the models are good at this because we can just set them loose and they can permute through their training and because a lot of things are common patterns, this can work, but is also generating massive amounts of slop that people are dumping on maintainers who are getting overwhelmed by it. I really want to know how many failed iterations a successful run cost.

One unanswered question is if this is economical. Like, if this is just gobs of money and is unsustainable, what did we just spend the last few years doing? I think these things have value, but not like this.

building great software has not been constrained on throughput nor technical for a while. it's limited by the human aspect, and there is no reason imo to believe by default that ai would change this.

I don't want to toot my own horn but I developed a mix of Photoshop / Pixelmator for AI assisted Photo editing and I use it every day instead of Pixelmator Pro [ 1 ]. I think it is in many respects the better app because I added many features I always wanted to have (replicate layers, any content as masks, crazy effects, etc). Granted it is buggy and still needs polish, but most software has bugs. So I've completely switched over to a complex AI generated app.

[ 1 ] https://pixellect.app/