# How To Build a Dark Software Factory in 3 Months

Channel: Craft vs Cruft
Video: https://www.youtube.com/watch?v=9wET7FyTo5I
Duration: 9 min
Language: English
Words: 1508
Transcript page: https://viewrankai.com/tools/youtube-transcript/9wET7FyTo5I

---

[0:00] All right, you want to talk about it? Let's talk about it. I'm Ray Myers, welcome to Craft versus Craft. And today we're going to talk about the dark software factory. If, like me, you're an expert in platform engineering or coding agents, you may have been getting the questions I've been getting. What do you think about dark software factory? How do you make a dark software factory? If not, I'll catch you up real quick. This is essentially the view that as we adopt coding agents, the final form for using them in software development is essentially to use only them. That they will take over the code completely. The

[0:37] dark in dark software factory is like the lights are off. We no longer see the code. Now, if you know where I'm at philosophically, you you may suspect this is not where I actually think everyone should go. But never mind that, I'm going to tell you the easiest way you're ever going to see of making a dark factory of your very own. In just 3 months. And not 3 months of work, but in 3 months to make one with almost no work. Don't believe me? Well, maybe you

[1:07] shouldn't believe everything you hear. But nonetheless, part of the reason that people are saying this is inevitably where things go is they have a a view of where language models and other technology fit in that is informed by this this destination, this kind of final destination called AGI. There's no unified definition of AGI, but one that I think is pretty significant today is the one in OpenAI's charter, their founding document, which is highly autonomous systems that outperform humans at most economically valuable work. So, in other words, it's a layoff machine. This is massive labor automation. The historically large funding rounds, the historically large IPOs that these companies like Anthropic

[1:56] and OpenAI are heading towards is related to the fact that there is a belief they can create hugely disruptive labor automation and therefore mass layoffs, mass unemployment unless we find some other way to make that happen, [clears throat] right? So, generally speaking, like that is why there is so much capital behind these companies right now and behind certain narratives of of how the tech is to be used. This creates tension if you're going about

[2:31] your business trying to engineer things. Put a pin in that for now. Now, on what schedule are we going to build the layoff machine? Well, with respect to software engineers, the layoff machine is actually very imminent. According to Dario Amodei, we might be 6 to 12 months away from models doing all of what software engineers do

[2:53] end-to-end. So, we're talking about strictly model capabilities and we're talking about all of software engineering end-to-end. Now, of course, I'm not quoting him to tell you that this is fact. I'm just saying that is the position of Anthropic, one of the leading model and agent providers for coding. So, taking the outside of that 6 to 12

[3:16] months, we're talking about December. Just another 3 months or so. All right. So, here's how to build a dark factory in 3 months. Step one, take Dario seriously. This may be the hardest part for some of you. Step two, wait 3 months. Now that it's December, prompt Claude. Say, "Please make a dark factory. Make no mistakes."

[3:42] Done. There's your dark factory. Right? Because anything you would have had to do to set up a dark software factory would have been software engineering. So, if that has been automated end to end, DSC, I should not have to help you make a dark software factory. It should make itself if what's promised can actually be

[4:06] delivered, right? So, why, if that can happen in sometime in the next 3 months, why would I do a lot of work trying to make one now? Why does no one else say that the way you make a dark software factory is you wait 3 months and then it builds itself? Why does no one say that with a straight face? Why am I only

[4:28] saying that because I am joking? Well, it have to be that the model labs are telling world leaders, investors, and executives things that software engineers do not find credible, wouldn't it? And I think that's really problematic. I don't want my dev tool vendors to be falsely announcing the end of my field. I find that to be

[4:51] counterproductive. The agent platform company Tesel describes the dark factory as the end state of loop engineering and real for a handful of elite teams, contested as a general model. That sounds about right. One of the main elite teams they refer to were involved in in breaking out the the term, which is strong DM. Some of the techniques they advertise make their dark factory work, right? They have digital twin universes. They have gene transfusion. They have all this This sounds like a bunch of engineering that they did. This doesn't sound like the models just figured it out for them. And you're going to notice that as a pattern

[5:29] generally when someone's got an idea about how you might build a dark factory, there's a whole lot of weight to the solution. It is nowhere near agents are just sort of ready to take over. It's a whole lot of add more agents to fix the problems the last agents introduced and and so on and so on and so on. Tesla in fact quotes them as saying you ought to be burning a thousand dollars in agent tokens per

[5:53] engineer per day. Um okay, is that is that cost-savings? I don't know. So this is a lot of effort for which the benefit would have to be really extraordinary, I would think. And you know, may maybe you could argue that it is. Uh that's that's another conversation. There's another project that's been out for a while called N Wave. I've heard a lot of people like it. This is in the family of thing that uh claims to be able to really gear together in a in a more or less lights off way. You know, this is coming pre-packaged with dozens of agents, hundreds of skills, very complicated

[6:26] stuff around just the models and agents. Yep, maybe that's worth it. Maybe you finally get to prompts as code if you just add enough agents, and that's actually a good solution, a place worth doing all that work to get to. I don't happen to think so. I think compilers are really good compilers, and LLMs are good at other things. But today I'm not really here to make the technical argument either way on that and say which use cases they might apply to. I'm just pointing out that if the Model Lab narrative was what caused us to think we wanted a dark factory in the first place, the way we're building it is

[7:07] nothing like what the Model Labs are saying reality is right now. And I think that's a problem. I think there's a lot of cognitive dissonance here. So this is an argument from trust. What we're being asked to do is take one of the most inefficient components in all of software infrastructure, the foundational LLM, and become completely dependent on it at every link in the chain of software engineering, perhaps of all knowledge work. And at the same time, in order to sell that vision, the Model Labs have wound up saying things we find highly untrustworthy. So can I theoretically execute the steps that people are saying lead to advanced loop

[7:48] engineering and then a dark factory? Like, yeah, I I I think it can be done. You can end up with something roughly shaped like that, but I have just no confidence in the long-term outcome there. And from a professional ethics standpoint, I find it impossible to recommend. If you believe that in a particular situation doing something like this actually might be just fine, and you understand what society trusts us to do as software engineers, that our systems are secure, respect privacy, are maintainable, and so on and so forth, perhaps you might reach a different conclusion. But from where I stand, it's a pretty tall order to make that case

[8:29] the same month we're talking about OpenAI's benchmarks causing them to accidentally commit felonies by hacking other companies. Anyway, thank you for joining me. This is part of a series of content called Dark Factory Considered Harmful. I'll link some resources on professional ethics, labor organization, and some other material I put out about the future of software engineering. My talk, Code Worth Writing, from Software Should Work, which is already out, and I'm working on talks for these other two conferences later this year. Thanks for watching, and as the Daoist says, with patience, the most tangled cord may be

[9:04] undone.

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt