Company Distillation

I make the case that after figuring out the Software Factory, a short next step will be complete distillation of products and by extension full companies, Company Distillation, as their margins and bloated structures come under threat.

AI will destroy software! We've been hearing that headline, software developers have been worried about how AI can do their job better and better, and will their work exist? I don't think it will. At least the job won't look like how software is built today. AI can do the job at maybe 80% of the quality of normal engineers, this is debatable of course, but what is not debatable is it is orders of magnitude cheaper at generating code. More code requires more supporting systems for that code, Quality Assurance (QA), continuous integration (CI) systems, browsers/specialized environments to test the code, scaling version control and merging/review processes to deal with a lot more code. And more!

I think it's clear the software factory will work in some form, pretty soon, maybe it already does. I think it does, as I already have it working for me! Agent Starbase is 100% agent written, increasingly AI tested (10% perhaps) and growing, and I think I'll get to AI generating requirements + features I actually want to add in the next few months. But even if I don't figure out the last thing, someone else might! It's becoming clear, if we have Software Factories/Dark Factories, expensive powerful software products that are worth billions a year will be at risk to lose some of their revenue + profitability to companies with working dark factories. I call this Company Distillation, this is borrowed from model distillation, where models are "copied" by competing model companies by paying a lot of money for receiving model outputs directly from the model via the API and then training a new model, often smaller, on those outputs.

It's Happening

We've been seeing the first signs of this, Adobe Photoshop clones, Office Suite clones. This type of software is a huge part of the market. This software will be easier and easier to clone with AI. Right now the software doesn't really work and sometimes this cloning process has just been straight IP theft, but I think we're seeing the start of something most companies are going to have to worry about. If you don't disrupt your own business to make it more automated and efficient and more useful to your customers - you could come in for a shocking surprise. The question now isn't if it's possible, but when it's possible.

The example that comes to my mind is Pages (the Microsoft Word equivalent) from Apple is able to open .docx documents. PDF files invented by Adobe can be opened in any browser and a variety of programs. Since these formats don't have IP protection, there is little in applications that can be protected (I'm not a lawyer, not legal advice). Even if it's not legal to use Reverse Engineer Anything to generate clones of applications, some version of clean-room implementation is already doable by agents. It's complicated by the fact that in their training data they might already have knowledge about a variety of profitable software products in the weights. But I think it's as defensible as any other code written by LLMs if the process is documented clearly and done correctly so it can later be defended in court.

Here's a process I imagine would work.

Buy/license the product legally. Document in depth how it works, either via human or agent usage. Put those requirements into a detailed folder structure of requirements. Make the documentation extensive, every input, behavior is documented. Real data and inputs and outputs are tested and their output behavior recorded. Have a human look through the requirements to make sure they look okay. I would imagine the new company would also want to improve on the previous product as they can be hard to use, bloated, have legacy implementations, complicated features that are hard to support + build. This software doesn't need to be focused on 100% of the users of the software, mostly the budget conscious < 50% who don't have long procurement processes and niche uses that the old software supports.

The requirements all get given to the software factory and the software factory works one to four weeks on the product with a built in QA step as it's going, and perhaps human engineering. The processes/code quality goals/automated measurement of software quality need to be well specified up front. The code needs to be 100% covered on the UI/backend/database. After it's built, there will have to be months of refining, then user testing, working out all the bugs/kinks. This doesn't include sales/marketing of this new software. But even imagine this takes 6 months to do with a $1 million budget with 10 people, there is almost no software company in the world with these costs. The price could be five times less expensive quite easily and still deliver unbelievably high margins.

Conclusion

I think this must be terrifying for any big company that exists, maybe any company at all. Other people can clone your product without the years or decades of development that you spent on it. Maybe they can even clone hundreds of products like yours with a small team! You're a sitting duck, just waiting for it to happen to you. It's important to get ahead of it.

If you are at a big company there is a lot more opportunity now too! You can develop software cheaply! Fewer people can create new features, you can enter your competitor's niches quicker, and your product can be customized for a small subset of your users, who get their dream experience in your services. It's never been a better time to be a customer!

I think this wave of company distillation is coming quickly! I could be working on this now. But there is also incredible opportunity for everyone who's going to see the cost of their software, services, products go down as custom software makes things more efficient. There will also be inevitable disruption, but you can build anything now too. Any pain you have is fixable with a little creativity. Computers can actually help you do what you want without a specialized skill set. A small tool you build is a lot less work to maintain and you can combine a few of them to create a safety net in case your life is disrupted by the roaring wave of automation. A negative trend for big and profitable companies will be a positive trend for all of us.

Ride the wave.