The way software is consumed and priced has gone through many iterations over its history. (Not that) long ago, we had “Shareware” – a limited version of the product free to download, use, and share of course. It came along with the option to purchase and own a more expanded (and perpetual) version of the product (aka full license) if your needs demanded. As software has moved to be more cloud-native, it has largely shifted away from perpetual, per-device, licensing model in favor of “Freemium” models. You sign up for a subscription and get personal-use, or a limited set of features for free, and then purchase a subscription when you and/or your team outgrows it. This concept of feature scalability is what has made product-led growth possible, and prevalent.
The debut of AI’s capabilities in software applications has started to make things more interesting. The classic freemium pricing model doesn’t work so well anymore because anytime you leverage AI features in the product, you as the software developer/provider incur a compute cost. Developers are willing to provide software for free for the purposes of marketing, growth, and/or community goodwill… but not at a financial loss. So, now users find themselves with a limited trial period and then a paywall for nearly all software products since AI is becoming a core component of modern applications. This is quickly leading to subscription fatigue. There is plenty of software that I am personally willing to pay for… but not monthly… for everything. Trying to predict AI usage for users also leads to major complexity in trying to even appropriately price usage tiers of a product – but that’s a whole separate discussion in itself for another time.
As a software developer, using Apple’s Foundational Models Framework may end up to be a transformative way to deliver software with AI and cloud-native features that won’t incur an ongoing operating cost to scale. There are lots of amazing capabilities that AI could add, and applications that people would be willing to pay for, but just not necessarily on a monthly subscription basis. Being able to leverage the local on-device models for software products is actually transformational… and I truly think application developers will be able to get much more value out of it than just making e-mails more friendly or professional (aka the current primary use-case for Apple Intelligence built into the OS’s). Pair that with existing iCloud sync capabilities, and as long as your app lives in the Apple ecosystem you could deploy and sell an AI-enabled product that looks and feels cloud-native without the ongoing operating overhead and usage costs.
While I am mentioning Apple’s ecosystem here, since they tend to lead industry trends for app development…. It’s important to note, that the concept of on-device LLM and intelligence is not uniquely theirs. Some Android hardware/software pairings, and Microsoft Windows 11 on capable PC’s are also making similar leaps to expose on-device AI capabilities to developers.
Ultimately, I think we will start to see the capabilities of AI become somewhat generally stratified so everyone will have some general clarity on what types of things are possible on-device, freely in the cloud, for a price in the cloud, and for a premium in the cloud. I have read multiple posts from people who generally point to local models as a giant waste of time and not something anyone should put efforts behind. Because? They have limited capabilities, and the cost of public compute will continue to drop closer to zero. However…. I think the big picture is getting missed here. Close to free, isn’t free. And the more it scales (usually the goal), the further from zero it becomes. Also, we are all carrying around devices in every form-factor that have way more compute then would have dreamed possible even a few years ago (and increasing every year)… most of which is sitting there doing nothing most of the time. And, believe it or not you don’t have to be running the latest huge parameter models to do lots of helpful things with AI.
I really look forward to seeing what the next generation of AI software looks like when developers are able to start leveraging local on-device models to add increasingly rich capabilities to apps we use every day!
