Let’s be honest.
We don’t write code to suffer — we write it to ship things that work.
As developers, we already use a ton of tools to be faster and more effective. Android Studio, IntelliJ IDEA, linters, static analysis, build systems… nobody questions those. AI is just another tool in that same category. A powerful one.
So why use AI during development?
Because it saves time.
And because time is expensive.
AI helps me move faster, especially when I’m working on something unfamiliar. Exploring a new project, digging through a legacy codebase, or switching context between systems is mentally costly. AI reduces that friction by helping with research, generating alternative approaches, reviewing my own solutions, and cutting down the time spent jumping between documentation and code references.
Context switching is expensive. AI helps me stay focused.
And yes — I really don’t enjoy writing unit tests. That’s a perfect job for my AI buddy. It’s also surprisingly good at suggesting edge cases I might overlook and helping with documentation for things I already understand but don’t feel like explaining in writing.
What Is Codex?
Codex is an AI-powered tool that runs directly in your terminal.
That alone makes a big difference.
Instead of pasting snippets into a chat window, Codex can actually work with your local codebase. It helps you understand what already exists — the structure, patterns, and intent behind the code.
This is where Codex really shines. It’s not just about “what code should I write next?” but rather “what is already here and why does it exist?”
For legacy projects, this is gold. Large codebases often lack documentation, contain historical decisions, and accumulate accidental complexity over time. Codex significantly reduces the time needed to research and orient yourself in that chaos.
Why I Like Having AI in My Terminal
When working on legacy systems, context is everything.
And legacy systems are usually everything except clear.
They are big, poorly documented, and full of hidden dependencies. Having AI in the terminal lets me stay in my workflow without constantly switching to a browser or IDE panels. I can ask questions about the entire project, not just a single file, and quickly understand how different parts relate to each other.
It feels less like autocomplete and more like having a quiet teammate who already scanned the repository.
Example from Practice (aka: Learn From My Mistakes)
I was configuring R8 and dealing with ProGuard rules. We already had them — or so I thought.
Turns out the rules were placed in a different directory. R8 didn’t see them at all.
The result?
Hours of staring at the same error message. Growing frustration. Growing anger. That special feeling when you’re convinced you just broke everything and your code is now cursed.
I wrote a lot of code, nothing worked, and I was genuinely nervous.
Codex helped me quickly locate the existing ProGuard rules, understand where they lived, and realize why R8 completely ignored them. The problem wasn’t my logic — it was configuration and structure.
I was mostly angry that I didn’t use Codex sooner.
Terminal AI vs IDE AI
IDE-based AI tools are great. They excel at autocomplete, small refactors, and inline explanations. I don’t use them myself, but I completely understand why someone might.
But terminal AI tools like Codex operate on a different level. They are much better for understanding entire systems, answering architectural questions, researching legacy code, and helping with scripting or automation.
If I had to simplify it: IDE AI works on the micro level — files and functions. Terminal AI operates on the macro level, encompassing modules, systems, and architecture.
They complement each other, but for exploration and understanding, terminal AI feels more powerful.
AI Is Not a Replacement for Engineering Discipline
Here’s the important part: don’t be a fool.
AI is a great tool, but it can:
- hallucinate
- misunderstand context
- confidently give you wrong answers
It doesn’t replace responsibility.
That’s why traditional tools still matter. I really enjoy static analysis tools like Detekt. They objectively detect complexity, point out code smells, optimize imports, and enforce coding standards.
Static analysis is authoritative.
AI is supportive.
Use both.
Limitations You Should Respect
AI tools — including Codex — have limits. Large codebases can exceed context windows. Security-sensitive code requires extra caution. Generated suggestions must always be reviewed and tested.
AI should assist your thinking, not replace it.
Final Mindset
AI makes you faster.
An engineering discipline makes you safe.
The best results are achieved by combining AI for research and acceleration, static analysis tools for quality assurance, and tests and reviews for correctness.
Used correctly, Codex becomes a powerful companion — not a shortcut.
Leave a Reply