AI killed software development, and I don't mean that AI will generate your code. Quite the contrary, we've got more than enough code to last us a thousand years, and having AI create software might work for some of us - But that's not how AI will kill software development. Quite the contrary, AI will kill software development by making code "obsolete", not by generating thousands of lines of code based upon some prompt created by a non-software developer. To understand why, you must understand the power of meta data and LLMs.
Meta data
Your database is exposing "meta data". Such meta data allows Magic to retrieve information about its table and columns, etc. This again can be used to scaffold Hyperlambda Workflows, that can be wrapped into an AI agent. The end result becomes an AI agent that allows you to manage your database using nothing but natural language. Watch me create a CRUD-based AI agent in some few minutes in the following video to understand what I'm talking about.
The resources required to build something such as the above and deliver it as a traditional GUI-based app, would easily imply $50,000 or more some 2 years ago. In the above video I'm creating it in some 5 minutes using our Low-Code and No-Code platform, and the end result is orders of magnitudes "better" than whatever I could have created using traditional programming some few years ago.
This is because Hyperlambda is a "meta programming language" that allows for generating code based upon meta data.
CRUD Magic
CRUD means Create, Read, Update, and Delete, and is the axiom around which your database evolves. If you've got CRUD, at least in theory, you've got everything required to "handle your data".
Magic Cloud can automatically generate CRUD endpoints wrapping your database. Combined with the ability of using these endpoints as AI functions, this allows you to deliver fully fledged AI agents, managing all CRUD operations towards your database. Something that's illustrated in the above video where we're wrapping a database with dozens of tables into an AI agent.
By combining CRUD with functions such as "Scrape website", "Search the web", "Send email", etc - You can basically deliver a CRUD-based AI agent that manages all your data, in addition to executing tasks on your behalf. The end result becomes an "AI app" that allows you to automate everything you previously had to use a traditional app to do.
RAG
RAG means Retrieval Augmented Generation, and is the process of using VSS search, combined with OpenAI's embeddings API, to retrieve "context data". Such context data is associated with the prompt before the request is sent to OpenAI, allowing you to use context data as instructions to teach the LLM how to execute tasks. RAG can be used to associate verbs and subjects with AI functions, resulting in that if I write "create a new artist", then the RAG training snippet instructing OpenAI how to create a new artist is associated with my request.
This results in that OpenAI and ChatGPT understands how to invoke AI functions, that's performing CRUD operations towards the database wrapped inside of your AI agent. Additionally, other AI functions such as instructions for how to send an email, invoke some API, or access some sensor can also easily be added to the AI agent.
Natural Language Coding
ChatGPT and OpenAI then serves as a "bridge", transpiling natural language into a function invocation, that's securely executed in your Magic cloudlet. This allows you to use natural language to access your data. The end result being an AI agent you can ask questions such as for instance.
- Create a new artist
- Add another record for AC/DC called "Back in Black"
- Count how many records Aerosmith have published
- Etc ...
This is because of OpenAI's ability to transpile natural language into functions invocations, that's executed on a platform such as Magic Cloud, resulting in that the AI can actually do something - Making it become an AI agent instead of a simple AI chatbot simply regurgitating data we've supplied to it.
This has nothing to do with having the AI generate code. Quite the contrary, it never generates code, it simply invokes functions in some other 3rd party platform, allowing it to dynamically retrieve data on demand.
Automation on Steroids
The amount of automation this results in is almost impossible to understand for somebody not having seen it with their own eyes. It's almost like a personal assistant, always ready to do whatever you instruct it to do. Need to order an airline ticket online for Thursday going to a wedding in Georgia? With an intelligently built AI agent, this is just some simple prompt engineering - Easily within reach of anyone with some basic computer skills.
If you can formulate your request using natural language, and you've got a good AI agent foundation - You can have the AI agent solve almost any task at hand. Below you can see us using our website AI chatbot to literally "clean our house", by finding housemaid services in our local area.
In theory there's nothing preventing us from automating the entire above process, having the AI chatbot get quotes, for then to order services, pay the housemaid afterwards, and even review the service after the job has been done.
The amount of automation AI agents provides us with is simply sick!
The New Apps
The combination of the above technologies, allows us to deliver "AI apps" that does some pretty spectacular things. Below is a screenshot from a client of us who had us create an AI agent with access to every single company in Sweden, and their financial KPIs for the last 5 years.
It could be argued that the above AI agent is only a simple SQL database, combined with some prompt engineering and features, allowing the AI agent to dynamically retrieve data from your database as it needs it.
Facts are, AI agents are rapidly becoming "the new apps". This implies there will be an explosion of AI agents, solving every imaginable problem and wrapping every imaginable app from your ERP system and CRM system, to your headless CMS, allowing you to use generative AI to publish content.
At AINIRO we've already wrapped NetSuite, HubSpot, Spotify, WordPress, and almost anything you can imagine into such AI agents. And as long as we can somehow access your app's database or API, we can create an AI agent that interacts with it.
Similarly to how apps exploded about 15 years ago, creating an entire new vertical for companies - We at AINIRO believe this will result in a similar effect, where everyone will inevitably use different AI apps, similarly to how everybody started using apps on their phones as smart phones became a thing 15 years ago.
AI Killed the GUI
The Graphical User Interface (GUI) has been with us for almost 5 decades, but with AI agents based upon natural language, the era of the GUI is over. AI agents is simply a superior vehicle for delivering "apps", because of its availability and ease of use.
Whenever I log into for instance ClickUp or HubSpot, there's a million buttons I need to understand to be able to use it effectively. With an AI agent I only need to to remember a simple sentence being; "What can you do", and the AI agent will describe its features to me, suggesting tasks that it can help me with.
Not only does this result in a "self documenting AI agent", but it also provides a superior user interface, since the learning curve required to understand how to use the AI agent is non-existing.
AI Agents, the Future of Computing
10 years from now, most of us will smile a little bit when we think back at how the world was in 2024. We'll have AI agents we talk to, instructing these to perform tasks, that even today will seem like science fiction. In a way it can be argued that AI agents is the real computer revolution, because it allows everybody to "talk to a computer", where this was only available for software developers previously, using highly formal programming languages, which was difficult to learn and master.
With natural language, this obstacle has been completely removed, since now anyone can automate a computer process using nothing but natural language and plain English. In a way it can be argued ...
English is the new programming language!
And all of this is facilitated for the LLM's ability to understand natural language, and transpile English instructions into AI function invocations, that interacts with your database or your API, allowing you to use natural language to access your database.
AI Killed Software Development
If you've paid close attention to what I've written above, you'll rapidly realise that it's the end of software development as we know it. Previously, every automated process needed to be implemented and "coded" by a software developer. With AI agents this is no longer true. For instance, 80% of our code as enterprise software developers is handling database systems. When you can rapidly wrap all CRUD operations towards your database inside an AI agent, and you can declaratively add functions such as send email, scrape website, etc - The AI agent's destiny becomes to kill software development (as we know it).
2 years ago I had to create code to send an email with a report generated around my database. Today I can just tell my AI agent; "Create a report of our monthly KPIs and send to our CFO", and the AI agent will happily do it for me.
Facts are, AI killed software development as we know it. But it's our jobs as software developers to deliver AI agents that are taking software development out of its misery. It's our jobs as software developers to make sure software development as a profession ceases to exist. It's our job to create all AI functions required to have the AI snuff our jobs.
This implies that even though software development is dead, during the transition phase mountains of money and power will switch hands. A lot of this money will end up in the hands of the software developers who are the most capable of delivering AI agents, ensuring that personal incentives becomes the driving force in what changes the world.
This implies that even though AI is literally rug pulling every single software developer in the world today, it's also the most important opportunity we've ever seen through our entire lives. Billions of dollars will change hands, and a lot of it will go into the hands of those most capable of helping AI kill software development. This is the single most important event we've seen in the computing industry ever since we started using computers. This is the end of software development as we know it.
AI Killed Software Development, Long Live Software Development!
Top comments (4)
Software development is really a process of composition. You take pieces of things, whether statements in languages, library components, etc, and "compose" them all into a "solution". This is akin to monks writing manuscripts in monasteries.
AI applications are not "composed" in this sense, and so they are not really "decomposable" either. They can be trained, like say a dog, but they cannot be "programmed", "developed", re-assembled, or used piecemeal like software past. Like happened with the printing press, monasteries and monks are not needed for ai applications. I think this is what Thomas may ultimately be speaking to.
Yes, and that it will change - But not the way most people believe I suspect ...
I was looking for a simple analogy that could convey your point. The medieval manuscript monk meeting the printing press seems like one...
Lovely analogy :)