I have a vast network of software engineers, from former colleagues at major tech companies to developers at nimble startups. This access gave me what I thought was a solid understanding of how AI is being applied in the industry today.
My assumption was simple: AI is likely writing the majority of enterprise code, perhaps 50% or 60%, with that number set to climb through 2026. This doesn’t imply that developers are merely “vibe coding.” They are still architecting solutions, approving every change, and meticulously reviewing most lines of code.
However, whenever I shared this view publicly, a vocal portion of the audience expressed skepticism, believing AI was nowhere near that level of capability. So, I did what any engineer would do on a cold winter weekend: I created a poll to gather some real data.
The Survey’s Core Questions
I wanted to answer a few key questions:
- What percentage of both engineers and students are using AI?
- Are they using AI within their IDE, or are they opting for fully agentic solutions like Warp or Cloud Code?
- What types of tasks provide the most value from AI?
- For those not using AI, what are the reasons?
Over two days, I gathered responses from about 600 people, and the results were quite surprising.
The Shocking Reality: 95% Use AI
First and foremost, a staggering 95% of people across all disciplines claim to be using AI to write code. This number alone effectively dismantles the narrative that enterprise developers are not leveraging AI. But let’s dive deeper into that data.
How Students Are Using AI
Based on 166 responses from students, the findings were illuminating:
- 93% of students are using AI to write their code.
- The majority report feeling 5 to 10 times faster compared to writing code by hand. This isn’t entirely surprising, as many students are still developing their coding speed and proficiency.
- A vast majority, 84%, still prefer an integrated solution within their IDE, such as Cursor or Claude Code in VS Code, where they can presumably review the generated code.
This preference for IDE integration is a positive sign. It indicates that students aren’t fully sold on the “vibe coding” trend offered by platforms where plain text prompts generate entire applications. It seems these fully agentic solutions are more for non-coders, while real engineers want to see and control the code.
Students reported a confidence level of around 2.8 out of 5 in the code AI produces. Only two respondents felt confident enough to auto-approve and ship every change AI suggested.
So, what are they using it for? The top use case, by a large margin, was boilerplate code. This makes perfect sense. Setting up a new project and writing all the preliminary code just to spin up a simple app is a universal pain point.
The second most common use case was debugging, which was a bit more surprising. In complex, distributed systems environments, bugs are often either very simple to solve by eye or so intricate that AI can’t possibly resolve them. However, for students, it appears AI is a valuable debugging partner. I was also surprised to see “writing tests” ranked so low, which might suggest that students simply aren’t writing many tests—a task where AI can be incredibly helpful.
How Professional Engineers Are Using AI
Out of approximately 400 professional engineers who responded, the results were remarkably similar to the student data.
- 95% of engineers report using AI to write code.
- They reported almost the exact same speed boost and confidence level as students.
- The preference for IDE-based solutions over agentic ones was also nearly identical.
Where engineers differ slightly is in their application of AI. While boilerplate and debugging still hold the top two spots, the distribution is much closer across other tasks.
- Boilerplate Code
- Debugging
- New Features
- Writing Tests
“Writing tests” appears much more frequently, as expected in a professional context. “New features” sliding into the third spot also makes sense, as creating new functionality is a core part of an engineer’s daily work in an enterprise setting.
The 5% Holdouts: Why Not Use AI?
What about the 5% of people who don’t use AI? I collected free-form responses to understand their reasoning.
The number one reason was the belief that it would dull their coding skills. This is undeniably true. If you asked me to implement something from scratch today that I could have easily done a year ago by hand, I would be much slower. However, those who are heavily leaning into AI are making a calculated bet. They believe we will never return to a world where all code must be written manually, so they are willing to let that specific muscle atrophy slightly.
Other reasons revolved around a lack of trust in the AI’s output and the belief that they could simply write the code faster themselves.
There’s a valid point here, especially concerning the long-term effects of working with AI on a large, unfamiliar codebase. In the past, I became intimately familiar with codebases by manually reading, writing, and debugging code. That process builds a detailed mental map of the system—you know which classes to use, how they fit into the larger architecture, and how everything interacts.
With AI, it’s easy to ship features quickly, but you don’t build that same deep understanding of the codebase. It’s much harder to construct that mental map when you’re not the one navigating the command line and piecing things together yourself. These developers have a point, and it will be interesting to see the long-term consequences of this shift in two or three years.
A Mirror Image
Ultimately, the apples-to-apples comparison between students and professional engineers reveals almost the exact same patterns of adoption, usage, and sentiment. The industry has clearly embraced AI as a coding partner, even if a healthy dose of skepticism remains about its long-term impact on developer skills and system knowledge.