Riding the AI Storm: My Startup’s Wild 2-Year Journey Through Innovation and Chaos! (III/IV)
5 min readDec 9, 2023
An AI-Native World and Opportunities for Entrepreneurs
Disclaimer: These are just my musings, not gospel truths, so let’s have a fun debate!
So, here are my two cents on what an AI-native future would look like:
- The ultimate form of devices remains a debate: The next-gen hardware is like a promised sequel that’s taking its sweet time. Sam Altman’s Humane launched the AI pin in November, and many think it’s OpenAI’s crystal ball into future hardware. I’m not sold on that. Sure, OpenAI has hardware ambitions, and it’s highly likely Humane will join forces with OpenAI to roll out the real next-gen hardware. But for now, this AI pin seems more like OpenAI’s sneaky way to hoard more real-world image data. Why do I think so? 1. OpenAI’s past investments have been all about data hoarding; 2. Current large models are like clueless tourists in reality, needing a serious reality check (Rewind for more info). Anyway, ditching our smartphones isn’t happening anytime soon. Expect more hardware forms, different LLMs, different OSes.
- A few winners with Hardware + OS + Cloud combo: In the mobile internet era, OS differentiation was all about interaction (including hardware), brand, and ecosystem. In the AI era, it’s a whole new ball game with hardware, interaction, large model capabilities, ecosystem, brand, and computing power. Apple, Google, Meta, OpenAI, Microsoft — they’re all gearing up for their hardware + OS show. And let’s not forget the small but mighty players, possibly standing out with better data privacy or a more open ecosystem (where developers can pick and choose from all available APIs or open-source models).
- Middle layer opportunities: Today’s iOS and Android, plus a plethora of device models, are already a developer’s nightmare. The future, with its mix of hardware and large models, private and non-private deployments, is only going to get more hair-pulling. This is where the middle layer for developers comes in — compatible with different large models, unifying standards, one-stop development and deployment of Agents across platforms. Plus, there are those vertical Agent stores with different distribution logic, a middle layer bonanza for users.
- OS may resemble a messaging app: This is just my personal hunch about the AI-native interface. Imagine a bunch of AI agents on the left, including your personal AI, ready to chat like your best buddies. The chat interface could include typing, voice, voice calls, and even doodling. Doesn’t it sound a lot like WeChat? WeChat is basically an OS in disguise, just swap your friends with AI agents, and voilà, you’ve got the future of OS and Slack/ Teams, but the constant is messaging as the main way to boss around machines. Thinking about the ultimate form of the Agent market and distribution, it’s like WeChat’s friend list and search feature on steroids. If this pans out, Agents proactively messaging users could be a hoot (private channels as a main source of user acquisition in the AI era, haha).
- Existence of AI-Native Vertical Applications: Today’s platforms (like Amazon, Expedia) play the role of centralized sorting and recommendation. In the future, this will be replaced by personal AI assistants within the OS, and the barriers for apps will shift to supply chain and supply-side content. For example, ask your AI assistant to buy New Year gifts for friends, and it might fill your Amazon cart, then show you the cart for final confirmation; or various agents might push purchase messages in the OS chat box. Either way, it’s no longer about the UI and smarts of individual apps, but the supply chain. AI will further shorten the supply-to-demand path, allowing suppliers to train their agents to directly deliver goods and services to end-users.
Opportunities for Startups
Everything mentioned above is grand-scale speculation. For startups, there are loads of opportunities, like Gleen.ai in enterprise services; or a fully AI-native workbench OS for solo entrepreneurs, a Slack on steroids; or an Agent framework that integrates all major models, open-source or platform-based, for creation and deployment across different OSes and hardware; or better data storage and transfer solutions; or customized development for apps and individuals. This diagram neatly ties together some entrepreneurial opportunities I see:
- User information/data isolation and transfer between different LLMs and hardware, especially unstructured, unlabeled data.
- More user-friendly AI-native interfaces for developers to integrate their agents.
- AI capabilities beyond LLMs — virtual humans, text-to-image, video, real-time translation, voice, 3D, etc.
- AI applications or control panels that integrate human in the loop.
- Helping developers better discover and connect with supply-side service providers, and aiding supply-side in better development.
- Hosting various models and data platforms (LLM agnostic).
- In-depth Agent development tools, or simpler template-based tools — think of it as the Shopify for Agents.
- An ‘Amazon’ for various Agent components, including knowledge bases, APIs, workflow templates, virtual avatars, voices, prompts.
- AI social/content tracks.
- And so on…
In the last part, I’ll share some lessons learnt running a startup.
Please read through part II — IV here:
More about me:
- Product homepage: https://mindos.com
- YouTube: https://www.youtube.com/@Mind-OS
- Twitter: @KissonL
- LinkedIn: https://www.linkedin.com/in/kisson-songqi-lin-49309516/
- Email: k@mindverse.ai
