Products and Platforms
Highlights
- Generative artificial intelligence (GenAI) and natural language processing (NLP) advancements have been in progress for several decades, from the 1960s, when ELIZA was developed at MIT, to Google’s groundbreaking paper on transformer models in 2017.
- Industries and business functions have jumped onto the GenAI bandwagon and are actively looking for use cases where this magic wand can be waved.
- The implications of GenAI could go much beyond those of an accelerator or an automation agent—they can potentially change the entire SDLC process as we know it today.
On this page
Introduction
Industries and business functions have jumped onto the GenAI bandwagon and are actively looking for use cases where this magic wand can be waved. For IT, the high-value use cases are clearly in the areas of code generation, synthetic data, and the creation of artifacts such as test cases and scripts. Predictions from Gartner® for the future of GenAI use in their study mentions “By 2027, nearly 15% of new applications will be automatically generated by AI without a human in the loop. This is not happening at all today.”
Characteristics
Ready to use:
The lure of GenAI is that their solutions are like fast food - they come with various tasks that can be consumed straight out of the box. Leveraging pre-trained Large Language Models (LLMs), particularly in the realm of code generation, GenAl offers a seamless experience, with hyperscalers wrapping cloud solutions around them. These LLMs, trained on vast source code, can swiftly generate software based on plain English prompts. This approach radically differs from traditional code generation methods, which require intensive pre-processing efforts. These new entrants disrupt the LCNCIDT (low code no code and intelligent developer tools) market as GenAl initially manifests itself as intelligent assistants or copilots, enhancing the developer experience.
Re-imagined user interfaces:
Machines now comprehend human languages, marking a significant advancement from developers learning low-level to high-level languages for computer interaction. Future conversations may shift to voice-based, more human-like interactions with versatile user interfaces, altering development approaches. The future UI will be primarily chat-based, as end-users will no longer want to learn how to use a brand-new software interface when they can communicate with it like a regular person.
Today, developers utilize various integrated development environments (IDEs) for coding in languages such as Java, .NET, and Python alongside multiple tools for testing, packaging, and deploying applications, with chat-based interfaces augmenting SDLC touchpoints. Low-code platforms should, therefore, stay ahead of the curve in driving these innovations in the way tomorrow's applications are created and consumed.
Empowerment:
The biggest excitement around GenAI stems from the fact that it puts the power to create into the hands of the users. The grail is for citizen developers to be able to develop complete working applications and host them independently.
- LCNC platforms should take the lead in making this democratization of application development a reality. This can help organizations enable "Shadow IT" teams to bridge the crunch in developer talent.
- Developers look forward to leveraging GenAl's automated documentation features, encompassing a range of document types like system specifications, design specs, code documentation, test plans, and more, streamlining their documentation processes.
- GenAI's ability to understand and summarize existing code into easily understandable documentation is equally promising.
- Content generation (images, video, music) has been revolutionized in terms of speed, personalization, and independence from specialized creators.
Creativity:
Traditionally, good, old-fashioned AI (GOFAI) and ML techniques were good at classification (sorting) and regression (making connections). With GenAI,they have started generation (text, sound, images, and video). This is a paradigm shift, even for many AI experts who had maintained that creativity would be the final frontier for AI until recently.
GenAI is already being applied to generating and aiding designs in the automobile, manufacturing, and healthcare industries. It is expected that nearly 60% of design efforts for building new websites and mobile apps can be performed by GenAI. The key role of GenAI in application and test case design would be to quickly generate multiple options during the ideation process to assist the designer.
GenAI implications
Some of the capabilities that are being explored include:
- Writing code: It can generate code based on instructions in natural language or provide suggestions with code snippets on completing incomplete code.
- Translating code: One early application is for the transpilation and refactoring legacy code for application modernization.
- Explaining code: It can help developers by explaining what a selected piece of code does. This could be extended to build an entire knowledge base by reverse engineering the screen layouts, flow diagrams/ charts, functional/ technical documentation, and business rules.
- Optimizing code: Analyze code and suggest code vulnerabilities or even bug fixes.
- Beautifying code: Automatically customize the look and feel of screens/ applications by applying standard style sheets based on enterprise/ brand design guidelines.
Risk mitigation
While GenAl holds promising potential, it poses evolving risks such as factual inaccuracies, biases, and security concerns. With AI-generated code, quality may suffer, which in turn leads to increased technical debt. Service providers address these issues with citations and contextualization. Implementing GenAl responsibly in enterprises requires pragmatic approaches and guardrails.
LCNC platforms can leverage traditional code generation practices and GenAl models to ensure usability, correctness, security, performance, maintainability, and compliance. Key strategies include facilitating access to pre-trained LLMs, integrating open-source models like Hugging Face, and enhancing results through fine-tuning, retrieval augmented generation technique (RAG), and knowledge integration.
