AI Tools 2025

2025 was the year when AI coding agent tools took over.

My AI Journey in 2025

I started with Ollama¹, moved to LM Studio², Trae³, GitHub Copilot⁴, Cursor⁵, and, later settled with Claude Code⁶, and at the end of 2025 started exploring Google Antigravity⁷ as long as it keeps the free version with a generous weekly limit :)

What I use AI tools For

Every AI tool try to help engineers to assist in there work.
I mainly used it for Software engineering & development, and learning about molecular biology and genetics.

The Learning

I used the various combonations of AI tool to get help on my learning development. And based on how I progress with AI tooling and mental model, I tend to use Cladue code more, but I reach the current state after many attempts and unlearning the old habits and shifting the mental model to finally be comfortable with the Claude Code(possibly as it just works and literally infinite personalization possibilities). I still use other tools and explore new tools available.

My Advice on AI Tooling

Do I have advice on how to use AI tooling? Hell Yes possibly explore all, try to build, learn as you need, and make the mental shift for the AI tooling, and later choose what best fits for you and keep exploring as a habit.

Getting into software engineering means we are always learning and exploring new things.

My Current Claude Code Setup

1. Use CLAUDE.md to guide Claude to follow the guidelines for development. You also update it based on your experience where you see Claude make different choices.
2. I used to prepare the plan document for the individual feature/task and reiterate to refine it and later provide the plan file path to Claude and ask for the implementation. As Claude improved or my experience with it’s features, I moved to Claude Plan mode feature and later to one shot implementation, and I still store the plan document.
3. I love the Claude subagent feature, and it’s part of my everyday work. I setup it up as per the individual project I work inside the project root .claude/agents/ folder. Subagents that I use: code-standard, code-build, code-review, code-verify
4. I keep the manual update for the Claude permission for the individual project under .claude/settings.json
5. I use the MCP server with Claude code as the project required (Most used local MCP server)
6. I use Claude hooks to provide feedback to Claude so that it can improve on what it’s produced. Some hooks I am exploring and will continue in 2026: PostToolUse and PreCompact.

Looking Forward

There are many other settings that are under scanner as I explore and see advantages, I would be updating these in the follow-up review post.