It's already a couple years ago now since the first explosion of AI-powered developer tools, making it possible to build full apps with AI. VS Code-style editors like Cursor and Windsurf brought large language models (LLMs) from the likes of Open AI and Anthropic directly into the development workflow.

LLMs were wedged into dev environments, giving them full access to our codebases. Despite the UI feeling a bit clunky, it was a game changer! With complete project context via IDEs, agents could reason across files, run commands, and build features with minimal human input.

Claude Code, Anthropic’s agentic coding tool that lives in your terminal and helps you turn ideas into code faster than ever before.

Do we even need IDEs?

Today though, as LLM code output has improved in quality and reliability (especially with Claude Opus 4.5), AI-driven IDEs seem to be shifting the experience away from code editing, and more towards agent management. For instance, in October 2025, Cursor introduced Multi-agents with their 'Agent Interface', and Google's Antigravity launched as an 'Agentic Development Platform':

Google Antigravity is our agentic development platform, evolving the IDE into the agent-first era.

In a similar vein, coding tools such as Claude Code and Aider are agentic development tools that live in your terminal. Maybe we're getting close to the point where we don't need even need to look at code any more:

Agentic Developer Platforms

I like Ben's tweet above, but I feel like AI Code IDEs were a necessary phase, despite them never feeling too comfortable usability wise.

Back in 2024, Aider was probably one of the first practical terminal-based agentic coders, and pairing it with an IDE made it easier to see what was going on - especially with earlier models hallucinating.

Here's a video using Aider in the terminal panel of an IDE:

Human in the loop

This approach is still useful today, so agentic code tools like Claude Code easily integrate with existing IDEs. Povilas Korop highlights that just being able to see the filetree makes things more comfortable:

Given that AI can still make a real mess, an IDE is still really useful, but it can feel unnecessary and heavy when the AI is doing it's job well.

The Opus 4.5 Boost

Personally, I've found as models improved with every Claude release up to Opus 4.5, I code less and less. My Cursor Year highlights those changing habits: fewer editor tabs, more terminal-driven workflows, and long stretches where I didn’t look at the code at all.

Today I don't really code at all. When Opus can do months of human work in a few hours, there's no reason to. It's cheaper to pay!

Hi Claude Code,
Bye Cursor Invoice

Given you can use Claude Code in any interface - IDE, Terminal, or Web, it makes sense for me to give Claude Code, especially when it comes to cost. IDEs such as Cursor and Windsurf usage gets expensive. On top of the pro tiers, I was paying an extra ~$10 a day:

CLI-first Agentic Coding Workflows

Furthermore, on X, more and more developers are sharing workflows that show how multiple agents and instruction files can be composed to work efficiently – including this example from the creator of Claude Code:

The idea of setting up the tasks for Opus and leaving them to run sounds good too. It would be great to do it from a phone while walking the dog like this fun project fro Pablo Stanley:

The Terminal as the New Chat Box

Maybe as a concept, this is kind of new, but it’s been a gradual shift. Tools like Aider started off in the terminal and included lists of short commands to change code and interact with AI, and as the models got better, commands slowly became more like full natural-language interactions like in Claude Code.

According to Martin Treiber at ikangai.com (despite reading like AI wrote his article), the CLI naturally emerged as a home for agentic coding tools. Terminals have unrestricted access to dev utilities like git, compilers, and can surface outputs directly - giving both developers and AI visibility of tests and commands:

The renaissance of terminal-based development tools seems counterintuitive in an era of sophisticated IDEs. Yet nearly every breakthrough in agentic coding – Claude Code, Aider, Gemini CLI, OpenCode – launched as command-line tools first. This wasn't nostalgia. It was architecture.

Overall, I've found when using agentic coding tools, whether it's an IDE or the terminal, everything becomes a chatbox anyway. Bringing AI to the terminal brings natural language with it, and that changes how we interact with it, making it more accessible and possibly simpler than a full IDE.

This could also make it more familiar to non-developer users, which is why it's become a great tool for even more people. This article from Ben Tossell sums that up:

How I code with agents, without being 'technical'

Try it all

For now, I think combining tools is still the best way to go, and to do what works for you. But it's important to keep an eye out and try new methods. That's what I'm doing next with Claude Code, and Codex too!

Despite the hype though, there's still issues as shown in this thread: