Control your data

Capture what is yours

If you, the consumer, are confused by the pricing structure of the current AI/SI market, the token economy, then you have a blind spot. You are missing a valuable input, that input being data. There are a small number of levers that matter here: talent, compute, and data. And the industry is quickly running out of data (the industry is also running out of compute but that is a discussion for another day).

You are constantly producing data, start treating your work as such. Some of this data is obviously less valuable than others, but don't get too bogged down in that just yet. The data you produce every day has value. Everything you do creates data. You must start thinking about your life in this way. Moving forward, your value will increasingly come from the data and the quality of the data you produce. Everything is data, capture what is rightfully yours.

Every interaction you have with LLMs should be captured. You should own and control it. It is yours. It is not the property of the labs whose products you use, it is yours. Start treating it as such. You want to set yourself up so that you can take full advantage of this wave of technological innovation without giving up sovereignty over your data. This means creating and building on top of systems that allow you to capture, own and control the data you produce.

Subsidization?

Why are the labs willing to "subsidize" tokens through their subscriptions? Also, why did I put the subsidize in quotations? Let's answer both of those questions. First, they are more than willing to subsidize usage because they are getting something in return. They are getting your data, which, again, is extremely valuable. Why do you think SpaceX bought Cursor? Yes, they have a good product, yes, they have lots of talented engineers and an excellent enterprise sales team. But that is not why they paid $60B for it. They bought their data. That's where the value comes from. SpaceX determined that they must acquire coding data to improve their models, and there weren't many options with the necessary scale, Cursor being the only one they could realistically acquire. It's as simple as that. Treat your data the way SpaceX treated Cursor: as a valuable asset. That holds true for individuals and even more so for businesses. Your data is your moat. It is what makes you unique, value it as such.

Now, why did I put subsidize in quotations? To keep this short, I don't like the current discourse around token pricing and subsidization. People look at the consumer subscriptions and love to point to how much more value you get from them versus paying raw API rates. This misses some nuance in my opinion. The labs sell tokens at a fairly high margin. They must do this because raw cost to serve you those tokens is not their largest expense. The compute and the power it draws is just one of the costs required to run their business. Training the models themselves is very expensive as you likely know. But this cost structure does not explain everything. I cannot peer into their financials but I would assume the labs are not taking a loss on these subscriptions. I would find that hard to believe. Yes, they would make more money selling that same inference at API rates, but I wouldn't describe this dynamic as subsidization. Rather, I would describe this as opportunity cost. And they are more than willing to trade the opportunity to sell those tokens at a higher cost for the data they receive in return. It is an easy trade for them. Simple as that.

Detach from proprietary systems

You should value your data as much or even more than the labs do. Own it. And realize there is nothing stopping you from doing so. You can take full advantage of the lab's pricing structure while owning your data. But, in order to do this, you must detach from the systems they are trying to lock you into.

Don't get me wrong, I use proprietary applications from the labs extremely heavily. I am a power user of the ChatGPT/Codex desktop app. It's a great application. But, I am increasingly moving away from it and other proprietary products to take further ownership of my data. You should too.

Centralize your data

The guiding principle behind Woven Matter is centralizing your data. The app is built on top of a single SQLite database that captures everything you do. It saves everything in one place and allows every agent/harness you use inside it to access this data as you see fit. This is freeing in many ways. It removes lock in. You can work between harnesses, between different machines and different workspaces while keeping a copy of everything in one central place. Use this data however you like. Simply save it as an accumulation to take advantage of at a later date or take a more active approach and give all of your agents access to each other. Allow them to work seamlessly in tandem.

Eliminate switching costs

I do believe there are and will continue to be real advantages to using the proprietary harnesses from the frontier labs. Those mainly being the Codex and Claude Code CLIs. The labs iterate on these and train their models specifically to perform well in their own harnesses. Take advantage of that. Use them, but don't allow yourself to get locked into a single vendor. You want the freedom to switch at a moment’s notice and if all of your data is locked into one of their applications you lose the ability to do this. Use whatever harness you want (within reason, there really aren't many worth using at this point). Switch at any time without losing your work or your data. This is the vision I am building around and will continue iterating towards. I see a future of immense, rapid progress. To take full advantage you must eliminate switching costs. Don't allow yourself to miss out on that future.