Topics In Demand
Notification
New

No notification found.

Coding with AI: A New Shift, A Bigger Opportunity
Coding with AI: A New Shift, A Bigger Opportunity

42

2

In 2021, Copilot just completed your code. By 2025, AI builds entire apps, writes tests, and even deploys them. This is a giant leap. Are we witnessing the end of traditional coding — or the start of a more powerful way to create? Let’s unpack how we got here, what’s next!

 

As ML Models matured, several AI-powered coding assistants emerged during 2021-2023 — including Kite, Tabnine, and CodeWhisperer. However, it was GitHub Copilot that truly marked the turning point, becoming the most widely adopted tool and ushering in a new era of AI-assisted coding tool. It was AI pair programming editor trained on OpenAI Codex, various GitHub repositories, that suggests code in real time to help developers work faster and more efficiently.

The Pre-Copilot Struggle of programmers: -

Programmers spent a lot of time looking for workarounds of common problems conductive browser searches. It involved searching for similar codes for various problems that took time hindering innovation and optimization.

The Copilot Intervention:

Github Copilot offered code snippet suggestions & established patterns used in the most used frameworks/ libraries right in the IDE which can be checked with a simple shortcut which is a huge saving. Although, it was best for experienced programmers who can evaluate these suggestions or ignore them as they already know the context, logic and expected output. Non developers or inexperienced developers may struggle to code as it needs capability to effectively review and edit suggestions made by GitHub Copilot.

Agentification: Coding becomes collaborative

With the rise of powerful, multimodal large language models (LLMs), a new generation of AI coding assistants emerged — including tools like Cursor, Windsurf, Devin, and Replit. These tools went beyond simple code suggestions, offering a more immersive developer experience. They introduced agent-like capabilities that enabled even non-developers to turn ideas into fully functional applications — complete with testing, debugging, and deployment. They acted more like a team of developers and project manager together.

We looked at various AI based coding apps to understand the evolution across the years by comparing them based on various parameters spread across general features, SDLC value chain productivity and limitations.

Based on

Features

Then

Now

Type of tools

Apps/Coding tools

Kite, Codota, TabNine, GitHub Copilot, Microsoft IntelliCode, Alibaba Cloud Cosy and AIXcoder, AWS CodeWhisperer

Windsurf, Replit, Cursor, Deven

Context awareness

Line Based, 100-200 tokens

Project Level awareness, 200K+ tokens - 2 Mn tokes

Agentic behavior

No

Autonomous + One click deployment

Chat interaction

No

Planning + Decision agents

Code generation

Snippets

Multi-file generation

Customizability

Minimal

Very High

Collaboration features

None

Project Manager + Dev feel.

Memory of interaction

No

Long term memory+ earlier interactions and builds across prompts and iterations

Models

CodeBERT combine transformer-based architectures with code embeddings, LLMs Codex (based on GPT-3) etc

large-context LLMs (e.g. GPT-4-turbo 128k or Claude 2.1 200k)

Training

Github repository, Feedback from Users, Open-source code

permissively licensed code, Replit-native data, publicly available, GPT 4

 

SDLC Value Chain

SDLC Support

Not available

Available but with severe hallucinations. Ex: Instances of AI CX chatbots "inventing its own policy" and saying users that "generating code would be completing your work"

Team Collaboration

Not available

 

Languages

Python, JavaScript, Typescript, Ruby, HTML, SQL etc

JavaScript and Python code, but it supports 16 languages in total. The current list includes: Bash, C, C#, C++, CSS, Go, Java, JavaScript, HTML, PHP, Perl, Python, R, Ruby, Rust, and SQL.

Error Correction

No support, need to manually review code.

Can detect bugs in code writing in a snippet. But deficiency in optimized coding.         

 

Could create spaghetti code based on what it is trained on. Hallucinations may hinder manual review as well.

 

Limitations

1. Poor at Handling Large Codebases
2. Current Limitations of AI Coding Tools
3. Limited customization
4. No memory of previous interactions
5. Only solved boilerplate code
6. Lacked system-level capabilities (testing, deployment)

1. lacks architectural depth
2. Can generate messy, unmaintainable, spaghetti code
3. Struggles with multi-file context and dependencies
4. Large context windows are resource-intensive
5. Lacks security and trust
6. Tool calling is often default

 

From Snippets to Systems: AI Coding Tools Are Growing Up: -

  • These new tools didn’t just assist to generate snippets — they began to build. What began as autocompleting for functions is now evolving into system-level thinking.
  • AI isn’t just writing lines of code anymore — it’s engineering entire applications. Even a non-developer can now describe its aesthetic or emotion through prompts and these models can blend design with intend and create idea to app with previews along the way.
  • Choose the best-performing LLMs for specialized roles (planner, coder, tester, debugger)

How does it compare with Low Code No Code applications?

  1. Interface: Low code/No code applications are drag and drop UIs interface while AI powered tools utilize natural language prompts and has chat-based interaction.
  2. Customization: Low Code/No code is limited to predefined templates and components although AI tools can custom code in any language tailored to unique use cases.
  3. User Skill Requirement: Non developers or Business Users with minimal tech background can use low code/no code tools but AI coding tools are more geared towards developers and accessible for non-coders as well

The Opportunity: -

Faster Development Cycles: -

AI-assisted coding tools have unlocked a new way of building software — one that’s fast, iterative, and accessible. While they may require multiple prompts and refinements to get things right, they significantly reduce the time spent on boilerplate code, UI scaffolding, and documentation. This allows developers to move faster from idea to prototype. Across large codebases, they still struggle with coherence but there lies the underlying opportunity of collaboration between human intelligence and AI speed.

Rapid Prototyping & MVPs:

AI-assisted coding tools have lowered the barrier to building MVPs, reducing the need for a full development team in the early stages of product development. This has enabled quick launch of startups with lesser number of people.

Learning Opportunity: -

For non-developers, they could build a simple functional application or website just by discussing with AI. They could start learning not by writing syntax but by talking to AI, understanding the logic and learning by building.


That the contents of third-party articles/blogs published here on the website, and the interpretation of all information in the article/blogs such as data, maps, numbers, opinions etc. displayed in the article/blogs and views or the opinions expressed within the content are solely of the author's; and do not reflect the opinions and beliefs of NASSCOM or its affiliates in any manner. NASSCOM does not take any liability w.r.t. content in any manner and will not be liable in any manner whatsoever for any kind of liability arising out of any act, error or omission. The contents of third-party article/blogs published, are provided solely as convenience; and the presence of these articles/blogs should not, under any circumstances, be considered as an endorsement of the contents by NASSCOM in any manner; and if you chose to access these articles/blogs , you do so at your own risk.


images
Madhumay
Deputy Manager - Research

© Copyright nasscom. All Rights Reserved.