There is a lengthy debate, over 400 posts on reddit as of this writing, on the topic of usefulness of AI in software development on Experienced Developers reddit.
I read and analyzed all of them, so you don’t have to, and because this topic is near and dear to my heart. The software developer productivity is what my company does, so being in tune with developers’ attitude towards AI tools is very important to me.
I’ve noticed general skepticism on twitter/bluesky about AI – and from people who’ve really tried it. Also, I’ve seen thought leadership pieces from VCs like Sequoia who go on to clam we’d be better of educating the next billion developers, then working on AI tools.
But let’s dig in the current reddit debate as it is quite good. The Original Post is copied below – verbatim. Below it you will find a more complete analysis on the good and not-so-good use cases for AI in software development.
Overall, the consensus of the thread is that AI works best as an assistant rather than an autonomous coder. Even as an assistant it must be kept a close eye on. While some people seem bullish on AI tooling, there is very strong skepticism towards AI tools but, surprisingly so, sometimes even the skeptics acknowledge AI tooling usefulness in specific scenarios
One of the Big Bosses at the company I work for sent an email out recently saying every engineer must use AI tools to develop and analyze code. The implication being, if you don't, you are operating at a suboptimal level of performance. Or whatever.
I do use ChatGPT sometimes and find it moderately useful, but I think this email is specifically emphasizing in-editor code assist tools like Gitlab Duo (which we use) provides. I have tried these tools; they take a long time to generate code, and when they do the generated code is often wrong and seems to lack contextual awareness. If it does suggest something good, it's often so dead simple that I might as well have written it myself. I actually view reliance on these tools, in their current form, as a huge risk. Not only is the code generated of consistently poor quality, I worry this is training developers to turn off their brains and not reason about the impact of code they write.
But, I do accept the possibility that I'm not using the tools right (or not using the right tools). So, I'm curious if anyone here is actually getting a huge productivity bump from these tools? And if so, which ones and how do you use them?
Most Frequent Use Cases Where Developers Found AI Helpful
- Boilerplate Code Generation (Writing YAML files, API route patterns, class structures, and basic CRUD operations. Generating repetitive code like adapter methods, constructors, and ORM models.)
This one user seems to have hit them all! 😊
Auto-generating unit tests and test scaffolding
Generating READMEs, docstrings, and function explanations, summarizing code comments.
Writing SQL queries, bash scripts, and other automation scripts.
Spinning up a basic project with new frameworks and assisting with exploration in unfamiliar languages.
Code Refactoring (Simplifying or restructuring existing code and getting suggestions for improvements for readability and maintainability.
Most Frequent Cases Where AI Tools Were Not Helpful
- Incorrect or Misleading Code Generation. Often, the AI-generated code appears syntactically correct but is often logically flawed. Also,devs find it faster to write code themselves rather than debugging incorrect AI-generated code. This is the most dangerous flaw, this user explaining it well
Lack of Context Awareness. AI typically struggles with large, complex codebases and fails to understand dependencies.It generates code that often works in isolation but does not integrate well with the existing system.
Inefficiency in Multi-Step Refactoring. Devs report that AI fails to maintain consistency across large projects, requiring them to manually adjust AI-generated suggestions.
Poor Code Review & PR Analysis. AI-based PR reviewers like CodeRabbit generate too many false positives, making them less useful than traditional static analysis tools.
Redundant or Overhyped Use Cases. Many devs feel that AI is being over-promoted for tasks already covered by existing tools (e.g., IDE features, linters, static analysis tools).
Over-Reliance & Skill Degradation. Not super frequent, but some developers worry that using AI for simple tasks reduces their ability to think critically and problem-solve.
Conclusion
You’d be silly not to at least try some AI tools. The summary above can give you a good idea of what some good use cases are. I’ve talked to some people who tried AI early on, and on wrong use cases, and they were turned off right from the get-go. The reality is that this space is fast evolving, and you should be in the know.
As-is today, AI tools provide some productivity gains. However, they are not replacements for experienced developers. At Uno Platform we are investing in tools which make developers productive within their current environments, such as Hot Design. Also, we are keeping a close eye and thinking of these useful scenarios to apply AI to, as we don’t believe just adding a simple LLM to it will actually add value. So, stay tuned to our blogs and Dev.to account as there is more goodness coming on this topic.
Sasha
Top comments (1)
Nice overview, Sasha! There is definitely a solid amount of hype around AI, as with any new tech.