Agentic AI in Software Development

Agentic AI in Software Development
The future of digital jobs and software engineering with Agentic AI!
By 
kneeraj.com
Neeraj Kumar
Software Engineer & Product Manager building AI & SaaS products. Sharing my learnings on AI and product development here.
The Transformation Has Begun…
Software is eating the world, and AI is eating software
 The question isn't whether AI will transform your work—it's whether you'll be ready to harness its power.
The tools we use, the problems we solve, and the skills we need are all evolving at breakneck speed.
What Industry Leaders Are Saying
Satya Nadella, Microsoft CEO
"Nadella revealed that 20-30% of Microsoft's code was AI-generated"
Mark Zuckerburg, Facebook
"50% of Meta's software development will be done by AI within the next year"
Sam Altman, OpenAI CEO
"The rate of improvement in AI systems is faster than almost anyone predicted. We're entering an era where AI can genuinely augment human intelligence."
Top LLM Labs…
Top companies are making unprecedented investments in large language models and AI infrastructure. OpenAI, Anthropic, Google DeepMind, and Meta AI are collectively raising billions in funding to advance the frontier of what's possible.
$500B
OpenAI Valuation
$183B
Anthropic Valuation
500M+
Daily Users
ChatGPT reached this milestone in record time
About The Speaker
www.kneeraj.com
Neeraj Kumar
Software Engineer & Product Manager building AI & SaaS products. Sharing my learnings on AI and product development here.
Mr. Neeraj Kumar, BITS Pilani (Engg), ISB (MBA)
15+ Years Building Software…
I've spent over a decade and a half in the trenches of software and product development, crafting award-winning mobile applications and leading engineering teams at scale.
Currently, I'm building production AI agents at Oracle, Freshworks and BrowserStack, including sophisticated test case generator agents and self-healing test automation systems that are transforming how teams ship quality software.
What You'll Learn Today
LLMs &  Agents Evolution
Understand the fundamental differences between large language models and agentic AI systems, and when to use each approach.
How to build AI Agents : Agent Primitives/Building Blocks
Explore the core building blocks that power intelligent agents: reasoning, memory, tools, and planning capabilities.
Build a simple Code Editing Agent with Claude SDK
Master the critical skill of working effectively with AI agents—a capability that's quickly becoming non-negotiable in modern software development roles.
Phase 1: The LLM Revolution
From emergence to evolution…
The Birth of Modern LLMs
1
2017: Transformer Architecture
Google's "Attention Is All You Need" paper introduces the transformer model, revolutionising natural language processing
2
2018-2019: GPT & BERT
OpenAI releases GPT-2, demonstrating unprecedented text generation. Google unveils BERT for understanding context
3
2020: Scaling Laws
GPT-3 proves that larger models with more data unlock emergent capabilities never explicitly programmed
4
2022-2023: ChatGPT Era
Public access to conversational AI sparks global adoption and transforms how millions work daily
What Makes LLMs Very Powerful & Very Useful!
What changed from previous era AI models to todays AI models? What is the difference?
1
Generalisable to any task
Modern LLMs are trained on hundreds of billions of tokens from diverse sources—books, websites, code repositories, and scientific papers. This vast exposure enables them to understand context, nuance, and domain-specific knowledge across virtually any field.
2
Pre-trained, 0-short learning
Unlike traditional ML models that require task-specific training, LLMs can perform new tasks they've never explicitly been trained for. Ask them to write code, summarise research, or explain concepts—they adapt on the fly.
LLM Demos
Text Generation Demo
Translation Demo
Phase 2: The Age of Agents
When AI learned to act, not just respond..
LLMs vs Agentic AI: Differences
Large Language Models
Single Step for simple tasks
Executes one action or response per query, suitable for simple, direct tasks.
It can only Respond…
Primarily responds to prompts, generating output based on input without initiating complex sequences.
Agentic AI Systems (like human agents…)
Multi-Step : Works on complex tasks
Can break down goals, plan steps, and execute tasks with minimal human intervention
Can Take Diverse Actions
Executes complex workflows, uses tools, and iterates until objectives are achieved. This includes actions like reading files, writing files, calling APIs, executing code, and interacting with external systems.
Persistent/Has Memory
Maintains context, learns from past actions, and refines strategies over time
The Agent Architecture: Core Primitives
Planning
Breaking complex objectives into actionable subtasks and determining execution order
Memory
Short-term context and long-term knowledge storage for learning and adaptation
Tool Use
Ability to call external APIs, databases, and functions to accomplish real-world tasks
Reasoning
Evaluating outcomes, course-correcting, and deciding on next actions dynamically
These four primitives work together, enabling agents to operate autonomously whilst remaining aligned with user intent.
Claude Coding Agent
About Claude Code Agent
Claude Code is an autonomous coding agent capable of understanding complex development tasks and executing them independently.
Example Command
"Create a calculator app within the folder "calc_app_2" and write the code for doing calculation based on the input in cli. Write in Python".
Your Next Steps: Build your first Coding Agent OpenAI Agents SDK & Claude Agent SDK Today…
3 simple steps : (300 lines of code…)
Initialise Agent class which runs LLMs in loop
Provide Tool Definitions
Run the Agent
Example : Simple Coding Agent
The future of software development isn't about replacing engineers, it's about augmenting them. Master these tools now, and you'll be leading the transformation.