The Post GPT Software Era
We're entering a world of on-demand software generation at runtime. Apps that will be single use, highly personal, and instant. This is the era of ephemeral software, and it begins now.
👋 Hey, I’m Linus and welcome to an open edition of my weekly newsletter. Each week I write about all things AI. I cover your questions, write op-eds and create AI tutorials. Let’s dive in.
🤝 If you want to sponsor a future newsletter, and you have a product or service, you can reach 14.000+ creators, designers, business owners, operators, industry leaders that are interested in AI, tech and Design.
A brief introduction
In my post about what I think Apple is up to I alluded to single use software as being a thing. Since end of March, I’ve just seen more and more traces of this actually becoming a thing.
It starts with ChatGPT plugins, or even OpenAI’s Code Interpreter. I also curated a thread on whats currently happening to the internet as we know it. But it continues with more and more sophisticated agents and at run-time capabilities as shown in Greg Brockmans demonstration of GPT-4 a month ago.
With the above as our starting point we can actually start looking at what I think is about to happen and why the signals are worth keeping track of.
Ephemeral UI
The idea that most interfaces will get transformed back to chat (or command line) interfaces I think are very real. Most application UI in most apps are almost like cars, 95% of the life time of a car it’s parked and never used. Most UI is the same. The UI is there for us to use when we need it. Or it’s there to give us contextual awareness and handle cognitive loads.
But when, everything and anything becomes possible through natural language commands. The need for much of the UI we see and use today goes away. Or at least thats the idea.
Below is an example from Sean Grove - where he talks about ephemeral UI, and how we at runtime will have apps conjure from nothing and to then simple fade away. More or less exactly what I wrote in this piece. I really suggest you watch Seans demo below to give you a sense of what we talk about.
Single-use apps
If we have AI that at runtime get’s a request from a human to complete a task or a goal. That then gets to work, figures out what it needs to do, write a small backend and a front end, all in mere seconds. Then presents you with the UI necessary to complete the task. What will this mean for software development companies as we know them today?
→ Here is an example of a research assistant Chord.pub that makes research simple and automated.
I think first of all, it’s even hard for us to understand this new paradigm. Because we are used to plan, research and build out software to be robust, and tackle things at scale. We don’t build software for one person, or less one person with a unique problem. We’re looking for large TAMs’ and products that can scale to hundreds if not billions of customers.
This new paradigm on the other hand is completely opposite. AI will be able to build and run software for a problem with just one customer. And it does not matter that there is just one customer. The AI will be cheap enough and fast enough to cover any grounds. It will be groundbreaking many times over.
“Single-use apps will be a huge thing. If you need to solve a unique problem, and nobody has ever done software for that because not enough market. With an LLM even a problem with only one user, will be doable, enter your ask, and code gets written, problem gets solved. Runtime ends, app dies. Done. Single use apps are born.” – Linus Ekenstam 26th of March 2023
If no ask is to complicated or too small and anyone with access to an AI or AI agent can do this. What will happen to software companies? What type of software companies will be required at all? Will all software companies initially become API’s that AI simply consume? and then what? Will AI just bypass API’s and write it’s own proprietary integrations on the fly? Opening up this can of worms, produces a lot of questions.
If we take a quick look at the most horizontal products of today, think Notion, Typeform, AirTable, Trello etc, they have use-cases in so many industries and across many verticals. They might be positioned well in this new world of single-use apps. Where a lot of current single use software will be rendered completely obsolete. But highly integrated and horizontal products might stand a better chance at thriving in this new paradigm.
There are two time-frames to consider here. Short term and long term.
I think short term we will see one type of behaviour. Companies trying to stay relevant, build plugins for ChatGPT, make sure they have public API’s for the AI’s to use and consume, and figure out a way to stay relevant. Working and partnering together with the massive players in the industry. I would not be surprised if we see major collaboration between large brands and smaller and large brands. Think Google and Adobe, or OpenAI and 10.000 smaller startups (through Plugins for example).
Long term I think we will see the biggest shifts and changes. The fact that AI is here to stay, I think is obvious. Gartner is projecting the AI market to be a trillion dollar industry by 2030 ( I think it will be way more ). These larger shifts will cause a lot of causalities in the industry at large.
These bigger shifts and changes will become challenging for the software industry to adapt to. Customer behaviours will shift fast. If we also consider the fact that compute might become invisible, and that all we do is type or talk to different agents, the future of software will be even crazier. I do believe that some screens will be around, not sure which ones and why yet. But we are a very visual species. I do think it’s time to offramp the crocked neck and smartphones to something else. Be it a combination of devices with ephemeral AI, always there to help us out. Or one new device category to take the next couple of decades and take us from the smartphone to what ever comes after it.
Speaking of invisible compute. Below is a clip from last months global TED conference where Imran Chaudhri of Humane is openly showing off their device for the first time. I was slightly underwhelmed, but also impressed. It does look a lot like a version of OS1 from the movie HER. If you have not watched that, I suggest you do. I do think only talking out loud to your device won’t be great, since there are many times I want to type to be private. But as a companion that can reduce my interactions with a screen by 50% or more I think it’s incredible.
Software that will write software
With this new paradigm of AI, and AI agents. It’s not far-fetched to draw out a timeline and place a spot on that timeline that will mark the convergence of self-writing and maintained code. When AI writes software and that software also writes software. Just to help us. The notion that software is eating the world, will transform to AI eating software. In this new world software won’t directly be written by us and it most certainly will be very different from what we are used to today.
Today we have GPT4, PaLM 2, Github Co-pilot X, Hugging Face StarCoder and Replit Ghostwriter to mention a few of the LLM’s that can code. There are of course more. But just to illustrate.
These models are already causing ripple effects in the software industry. Both good and bad, but I’d argue mainly good.
Developers that are just starting their careers can in many cases compete with senior developers. The description of a developer will most likely change with AI being a tool, just like its currently relatively hard to define what a developer is or does due to it being a broad description of an entire field.
As we are getting closer to models that code on the fly, and they can handle larger and larger projects through a combination of larger context windows and indexed projects. One role of the developer will become more and more that of a curator with surgical precision. Be able to jump and switch between many systems and contexts.
While another one will be more that of an ideas person. Someone that is able to cleverly ask and form systems and come up with simple and powerful asks. Someone that is able to communicate well with the machine (almost like Neo in the Matrix) but with less hype. These new school developers will be in advantage if they also have a good understanding of how code works. But then again. I think lateral thinking and good written/spoken communication skill will matter more.
Once we are past that initial time period of humans and machines coding together. It feels more and more natural that we will allow the machines to write more and more of the code we use. Also makes sense that AI takes care of the quality assurance, and the bug and error handling, and continues improvement.
Just writing this sends chills down my spine, since in many cases these models are already black holes, what happens when much smarter versions of todays models also will write and maintain code at runtime. And more so, create single-use apps at run time that simple dissolves upon usage? There are so many questions, and so much that needs exploring that I think that humans will have their hands full, for quite some time trying to figure these things out.
I think it is safe to say that tomorrows world of software will look very different from today. The post GPT software era has already begun.
Linus Ekenstam – May 9th 2023
Thank you for reading, if you are still here below are some short tidbits from things I’ve encountered or done in the past week. Feel free to check them out. We got a new conversation with Jon, formless.ai and my next masterclass on Midjourney
AI Storytelling and Narrative conversation with Jon Radoff
I had the honours to talk to Jon Radoff that writes Metaverse Meditations here on Substack about the power of AI in Storytelling and creating narratives.
Formless by Typeform
I met with David Okuniev one of the founders of Typeform.com earlier this week, and he gave me a hands on view of what they have been working on at Typeform Labs… Formless.
I very interesting take on how you can have a casual conversation, with certain intents, and after the conversation has ended the unstructured data gets transformed into structured data.
It’s interesting to see Typeform disrupting themselves from the inside. Extremely excited to see where they manage to take this.
My second Live Workshop – Leverage Midjourney as a business
Workshop/Webinar on the 22nd of May, 7pm GMT+2
I'll be doing a ~2,5 hour online live MasterClass on how to use Midjourney in your business. No previous Midjourney experience required.
It’s first come, first served. $399
Still here?
If you are still here, thank you. I hope you enjoyed this muse into the post GPT software era. If you have questions, feel free to reply to this email or reach out to me on Twitter or reply/comment here on Substack.
Newsletter Sponsor opportunity
If you want to sponsor a future newsletter, and you have a product or service, you can reach 14.000+ creators, designers, business owners, operators, industry leaders that are interested in AI, tech and Design.
I’ve been thinking talking a lot lately with people about the “post app era“. Clearly I’m not the only one.
At some point, and more quickly than we expect, even the structure of the app itself is going to go the way of skeuomorphic design.
It’s not totally clear what will replace it but the age of the computer as AI biome is coming.
A thought-provoking and insightful article!
As the creator of a highly horizontal product I may be biased, but I also feel that such products are well positioned.
One reason why horizontal products are well positioned could be is that they already often serve as "hubs" for their user base, providing a central platform for managing diverse tasks and processes. This makes them an ideal foundation for AI-driven customization and automation, as they can easily integrate with AI technologies and adapt to the specific needs of different industries.
As an example, I can't think of a single feature in our platform where augmenting it with AI wouldn't lead to significant improvements in efficiency, user experience, and adaptability.