How Do Developers Interact with AI? An Exploratory Study on Modeling Developer Programming Behavior
This mixed-methods study of 76 developers introduces the S-IASE model to characterize the multi-dimensional nature of AI-assisted programming by integrating intentions, actions, tools, and emotions, revealing that AI usage fosters a more stable emotional flow and focused code generation while also eliciting complex psychological responses like impostor syndrome.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine software development as a complex cooking process. For decades, chefs (developers) have followed recipes, chopped ingredients, and managed stoves entirely on their own. But recently, a new kind of "smart sous-chef" (AI assistants like GitHub Copilot or ChatGPT) has entered the kitchen. These sous-chefs can instantly suggest recipes, chop vegetables, or even plate dishes.
The paper you provided is like a detailed observation of what happens when real chefs start using these smart sous-chefs. The researchers wanted to know: Does the chef just sit back and watch, or do they change how they cook? What are they thinking and feeling while the sous-chef works?
Here is a breakdown of their findings in simple terms:
1. The New "Recipe" for Cooking: The S-IASE Model
Before this study, researchers mostly watched chefs from the outside and counted how many times they chopped or stirred. They missed what was happening inside the chef's head.
The researchers created a new way to look at cooking, called the S-IASE Model. Instead of just watching the hands, they looked at four things happening at once:
- State: Where are we in the recipe right now? (e.g., "The sauce is burning" or "We need to find a spice.")
- Intention: What is the chef trying to do? (e.g., "I need to fix this error" or "I want to make this faster.")
- Action + Tool: What are they doing, and what tool are they using? (e.g., "Asking the sous-chef for a suggestion" vs. "Looking up a spice in a book.")
- Emotion: How does the chef feel? (e.g., "Happy," "Frustrated," or "Guilty.")
They found that when chefs use the smart sous-chef, all four of these things change and dance together in a complex way.
2. The "Trust but Verify" Dance
The researchers noticed a specific rhythm in how chefs used the AI. It wasn't just "ask and accept." It was more like a game of ping-pong:
- The chef asks the AI for a solution.
- The AI throws back a suggestion (a "ping").
- The chef immediately checks it, tweaks it, or runs a test (the "pong").
- If it works, they move on. If not, they ask again.
The Analogy: Imagine a chef who doesn't read the whole recipe book anymore. Instead, they ask the sous-chef, "How do I make this sauce?" The sous-chef says, "Add salt." The chef tastes it, says, "Too salty," and asks, "Okay, how about less salt?" They keep going back and forth until it's right. They are creating and verifying much faster, but they are spending less time deeply reading the instructions before starting.
3. The Emotional Rollercoaster (or Lack Thereof)
The study compared two groups of chefs:
- Group A: Used the smart sous-chef.
- Group B: Had to cook everything from scratch without help.
The Finding: The group using the AI had a much smoother emotional ride. They felt less stressed and less frustrated. They were happier and more confident.
- Why? Because the AI helped them get unstuck quickly.
- The Catch: Even though they were happier, some chefs felt a secret guilt. They felt like "impostors" (frauds). One chef said, "I finished the dish with the AI's help, but I feel guilty because I didn't learn how to make it myself." They worried that if they didn't do the work, they weren't really learning.
4. The "Show Me, Don't Tell Me" Problem
Sometimes, the smart sous-chef gives instructions that are technically correct but hard to follow because they are just text.
- The Scenario: The sous-chef says, "Click the button in the top right corner."
- The Problem: The chef can't find the button. The text is vague.
- The Reaction: The chef gives up on the AI, goes to a human forum (like Stack Overflow), or looks for a picture/diagram to see exactly where the button is.
- The Lesson: Chefs need visuals. They want to see the steps, not just read them. When the AI only gives text, the chef gets stuck and has to go find a picture elsewhere.
5. Who Uses the AI the Most?
The study found that chefs were most likely to trust the AI in two situations:
- When they were super familiar with the dish (they knew the basics, so they just wanted speed).
- When they were completely lost (they had no idea what to do, so they needed a starting point).
- The Middle Ground: If a chef knew a little bit about the dish but wasn't an expert, they were actually skeptical of the AI. They would ask the AI for help, but then double-check everything because they didn't fully trust the AI's answer yet.
Summary
The paper concludes that AI isn't just a tool that makes coding faster; it changes the entire personality of the developer.
- They become more focused on building and testing rather than reading and planning.
- They feel less stressed but sometimes more guilty about relying on help.
- They need visual aids because text-only instructions often fail them.
The researchers built this "S-IASE" model to help future AI designers understand that they aren't just building a code generator; they are building a partner that affects how developers think, feel, and work.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.