Why You Should Learn Coding in the AI Era
"ChatGPT writes all the code for you โ why bother learning to code?"
You've probably thought about this question at least once. In the lab, you can describe what you want to AI and get matching code. Paste an error message and get a fix. You barely need to memorize function names or syntax anymore.
But does that really mean learning to code is no longer necessary?
AI Is an Automated Pipette, Not a Protocol Designer
Let's use a lab analogy.
Just because automated pipettes exist, does a researcher no longer need to understand the principles of pipetting? Of course not. An automated pipette only handles repetitive dispensing tasks. Which reagent, how much, in what order, under what conditions โ this experiment design is still the researcher's job.
AI coding tools work the same way:
| What AI Does Well | What Researchers Must Do | |
|---|---|---|
| Role | Generates code and fixes errors | Designs what to build and validates results |
| Analogy | Automated pipette | Experiment designer |
| Limitation | Produces irrelevant code without context | Gets accurate code when given proper context |
The key point โ to give AI good instructions, you need to understand basic coding concepts.
What Happens When You Use AI Without Fundamentals
These are real situations that happen in labs:
Scenario 1: You can't understand the errors
You run code AI gave you and get an error. You paste the error back to AI and get a fix. Another error. Another paste. At some point the code is completely different from the original, and you have no idea what changed.
Someone who knows that TypeError: Cannot read property of undefined means "you tried to access a property on an undefined variable" can ask AI the right question. Someone who doesn't gets stuck in an infinite loop facing the same error.
Scenario 2: You can't verify if AI's code is correct
You run AI's analysis code and get results. But you can't tell if those results are right or wrong. Even if a for-loop has an off-by-one error that skips the last element, you just look at the output and think "seems about right."
Publishing AI-analyzed data in a paper without verification is like writing experiment results in a report without checking them.
Scenario 3: You can't write precise prompts
Saying "analyze this data with Python" gets you generic code. Saying "read this CSV with pandas, filter rows where the OD column is below 0.5, and calculate per-sample means using groupby" gets you precise code.
To write that second prompt, you need to know concepts like pandas, CSV, filtering, and groupby. Using AI effectively itself requires coding knowledge.
Three Powers That Coding Fundamentals Give You
1. The Eye to Verify AI's Output
Understanding variables, functions, loops, and conditionals lets you read AI-generated code and judge "this part is wrong." You don't need to write perfect code yourself, but the ability to read and understand code is essential.
2. The Vocabulary for Precise Prompts
Knowing coding terminology lets you make precise requests to AI. Someone who can say "make async API calls, parse the JSON, then map over the array" gets very different results from someone who says "just get me some data."
3. The Confidence to Work Without AI
Even if the internet goes down or the API is unavailable, you can write basic scripts yourself. Reading a CSV, filtering by condition, saving results โ you should be able to do this much without AI. And once you can, you'll use AI 10 times more effectively.
What Coding Means for Researchers
This isn't about becoming a professional developer.
| Goal | Required Level |
|---|---|
| Give AI precise instructions | Understand variables, functions, conditionals, loops |
| Read and modify analysis scripts | Above + file I/O, library usage |
| Understand a colleague's pipeline | Above + modules, error handling, API calls |
| Build your own automation tools | Above + frameworks, databases |
How far you go is your decision. But the first line โ "understand variables, functions, conditionals, loops" โ is a fundamental skill for researchers in 2026. Just as reading papers is expected, reading code has become part of research in this era.
When You Meet AI After Learning the Basics
The order matters.
AI first, basics later: "I don't understand what AI is doing, and I'm stuck when errors appear."
Basics first, AI later: "I know what I want, and AI builds it for me quickly."
DevBench is designed for the latter path. Understand what a variable is, what a function is, what a server is โ then be able to say "Build me a sample management API with Express, use RESTful routing, and add logging middleware." That is what real coding ability means in the AI era.