Soul of the New Apps

The apps that dominate our daily lives have been smoothly sliding LLMs into our dm-s. ChatGPT, Bard, Microsoft’s has Copilots, WhatsApp’s Meta AI, SnapChat My AI, X permium+’s Grok – while Alexa and Siri are now the boomers in the family. It would not be an overstatement to say we are witnessing a major paradigm shift in the soul of our app driven life style, and in this post I want to document my thoughts, observations and predictions on the changing landscape.

Conversation – The ultimate App UI

At the core of any human computer interaction we have, Input -> Compute -> Output. It has been a long road to reduce the friction at the boundaries of compute – making computers comprehend our intent and respond back meaningfully. First there were command line interfaces; then for decades we have been stuck iterating with the graphical user interface. In the absence of natural language comprehension, these were the only tools to narrow down the users expressions for computation. Take a look at “The Mother of all Demos” in which Douglas Engelbart demonstrated for the first time what a point and click interface would look like and ushered in the era of personal computing.

Fast forward to Steve Job’s 2007 demo of the first iPhone, was a giant leap that showed us a way to touch and manipulate a computation with our finger that actually usable. It introduced a new way to express our intent to the computer and we have been building on top of that for fifteen years.

Numerous attempts all along the way to make the computer understand natural language paved the way for what was coming in November 2022. A year after ChatGPT, we now have a functioning talking computer. LLMs can now see, hear and understand our natural language instructions. We no longer need to wrestle to make the results of computation comprehensible – we can get natural language description of the compute at the right level of compression that reduces the cognitive load for interpretation.
Here’s a screenshot of ChatGPT app on my phone, which would have been considered an impossible dream in 2010.

Every non-conversational apps like banks, brokerage firms, or utility company now are rushing to add LLM based chat bots on their sites.

Support chat windows on sites have been there for a long time, but they hardly worked either handing us over to a human or just giving up with a bad user experience. With LLM powered chatbots, we have entered a new era.

Jokes aside, my prediction is they will gradually move out of one corner of the screen and take over the whole UI relieving the users of pain of clicking through navigation menus and buttons, filling lengthy forms. All the GUI elements, which are really proxies for natural language, are going to melt away in the face of direct conversation.

Messaging apps that put conversation first, like Slack, Teams, Twitter, Whatsapp, Meta’s messenger have a huge advantage to build new platforms for businesses and industries. Back in the day businesses had only the physical platform to build brick and mortar shops. During the dot-com era, they moved to the internet when they realized the convenience of the customer. The next platforms were the App store and Play store as businesses realized customers were carrying a smart phone in their pocket and the desktop was just gathering dust. In 2024, we stand at the cusp of the GPT Store. Businesses need to build RAG (Retrieval Augmented Generation) apps and ditch the whole click driven UI for conversation first UI.
But RAG apps are just the beginning though. The reason a human to human conversation is magical is not only because we can recall information, but we can re-evaluate our positions and adapt to the changed state of the world. This takes me to the more wilder part of my prediction – just in time reprogramming.

Reprogramming – Apps Adapting to the world

Software has already eaten the world. But the world keeps changing. To keep up with the changes, software needs to change. Today, we have teams of developers doing elaborate design, build, test, ship cycles to build complex software. Business analysts, developers, testers, product managers collectively try to understand, implement and evolve the software to meet the demands of the changing world.

ChatGPTs Code Interpreter and Data Analysis have already shown us what just in time computing can look like. Stretch your imagination a little further where code generation, testing and execution all gets done in real time as response to your prompt. The fact that LLMs today can generate code at a speed acceptable to many real-time scenarios is the second big change in the soul of the machine. Niche business logic that earlier took months to code, test and ship could get autogenerated, compiled, tested and finally executed – on the fly based on a user prompt. Take a look at the End of Programming talk by Dr. Matt Welsh which concludes that today the “(Large Language) Model is the computer”.

My prediction is dynamic code generation and just in time execution with the continuous re-authoring of complex business logic based on user feedback will become how apps function in the future. As the landscape will go through tumultuous disruption, the pace at which we write code and ship software will no longer be sustainable. Current development best practices say all development, testing etc needs to be moved left. But as LLMs get better, we might see an absolute paradigm shift where the entire software development life cycle moves all the way to the right, towards the edge. The models will eat the whole software stack, not just code generation. The removal of humans from writing and maintaining code, will lead to evolution of the software stack to focus on quality, safety and security. “Conversation first & LLM inside” becomes the new stack.

The Future of our App centric life

From Gutenberg’s printing press, to the modern web and mobile App Stores – pivotal technology changes have given rise to platforms on which the next chapters of human civilization was written. Unfortunately in modern times pivotal technological changes have only widened the economic gaps between the rich and the poor. The promise of the trickle down economics remains a distant dream that never delivers as gig economy workers get strangled servicing multiple apps. It is evident that the success of the AI arms race is biased towards deep pockets, and the super wealthy tech giants have all the unfair advantage. Our only hope is that in the past, we have been successful in building open technologies that benefit the whole civilization as well. Take the Internet for example which triumphed over the proprietary Information Superhighway that Microsoft wanted to establish in the 90s. Open Source softwares like LAMP stack that got the world to its digital adolescence. We need open standards, protocols, weights, regulations and software for sustainable AI. That way the next generation of computing, it is not owned by a multibillion dollar corporation, but is level playing field that rewards our unique perspectives and helps us progress as a species.

Leave a Reply

Your email address will not be published. Required fields are marked *