AI in Application Development – Where Are We Headed?

Artificial intelligence is one of the hottest topics in technology right now.  It’s been an especially exciting time for developers in the last 12 months as the possibilities of using AI in application development expand.   

For example, Microsoft has released multiple new products and services that leverage AI or have extended existing products with new AI capabilities. 

AI capabilities can be developed using no-code, low-code, and pro-code offerings meaning all audiences are catered for.  Whether it be citizen developers, fusion teams, or professional developers. 

Fuelling much of these is the growth in natural processing, and in particular, large language models such as ChatGPT. 

Generative AI is Making Human and Machine Interaction More Intuitive 

Advancements in neural networks, natural language processing (NLP), and large language models (LLM) are making it easier for developers to interact with the machines they build on. These technologies are helping developers move more quickly from prototype to production and in fewer iterations. 

I’ve had my own “aha” moments when writing C# code with the support of GitHub Copilot and Visual Studio 2022.  I can say it has saved me time.   

One scenario involved fetching data from the web that was processed in multiple steps involving Azure Blob Storage containers. After adding a few code comments, Copilot wrote about 70% of the code to move the data through the relevant containers and in the correct sequence. 

Conversation-Layer-as-a-Service 

We’re seeing more products and services aimed at helping developers create conversation layers around existing datasets.  For example, Azure Open AI Service, ChatGPT, and Azure Cognitive Search can be used to create conversational interfaces that are grounded with your existing enterprise data, thereby guaranteeing the security and sovereignty of your data. 

Whilst still in nascent stages, we’re also seeing the infiltration of ChatGPT at the database layer in Azure SQL, thereby making it possible to query data using natural language.  

Advancements in speech-to-text and text-to-speech AI services are particularly exciting.   

These vastly reduce the barrier to entry for smaller development teams to create innovative voice solutions. With a small development team, it’s now possible to create a scalable “call centre in a box”. 

Where Are We Heading? What’s Next? 

Not long ago, the line between the target audience for AI tooling and services was clear, developers, and citizen developers. 

I personally felt that until recently, no-code and low-code tooling capabilities targeted at citizen developers made it hard to solve complex problems or create feature rich solutions. 

This has changed. 

We’ve already seen this happen with Azure Bot Service, Microsoft Bot Framework. Many of these capabilities are now surfaced in Power Virtual Agents, thereby making it easier for you to feature rich conversational AI solutions with an intuitive interface. 

Self Service 

We will continue to see the coalescing of capabilities of pro-code tooling into low-code and no-code tooling.   

We will see more capabilities being made to citizen developers in self-serve interfaces.  The recent Language Studio and Vision Studio UI tools are good examples of this.  These UI tools offer low-code, self-serve interfaces that let you create custom ML models with text analytics and computer vision capabilities.   

Most self-serve interfaces behave in a similar way in that they let you visually evaluate how an AI service operates.  After creating your ML model, the interface exposes an API endpoint that can be integrated with your wider software solution.  It can then be consumed using either a REST API endpoint or a free dedicated developer SDK. 

Increased Velocity 

I believe we’ll see an increase in fusion teams whereby citizen developers and developers will work much closer together and in parallel to achieve business outcomes.   

Innovations in generative AI will make it quicker to develop minimal viable products, demos, and software.  AI will continue to be integrated into the software development lifecycle and further improve the software development feedback loop.   

With increased adoption of AI, and consequently, the democratisation of AI model creation, I expect MLOps (Machine Learning Operations) to be another growth area. 

Copilots, not Autopilots 

The dust is settling and despite AI scare mongering, developers will still be required.  One of the biggest shifts is the concept of the copilot.  

 We should remember these new tools are copilots, and not auto-pilots. They exist to help augment human capabilities.   

It’s an exciting time for developers and the IT industry as whole.