As a technical program manager at Devblock, even though I’m not actively coding on projects, I’ve been diving into AI-powered code editors to see how they can enhance our development workflow. One tool that has caught my attention is called Cursor. It has streamlined aspects of coding for our engineers, offering smart suggestions, automating routine tasks, identifying issues and resolving errors, and analyzing code, freeing up our developers to focus on other critical aspects of the role such as ideation, collaboration and creative solutioning. The “agent mode” and “compose tab” features of Cursor that recently came out are impressive at the breadth and depth of code changes they’re able to effect, while giving the engineer clarity of and control over the results.

---

We’ve also been exploring other such tools like GitHub, Copilot and Tabnine. GitHub Copilot integrates seamlessly with VSCode, and Tabnine supports a wide range of IDEs. The list of AI-assisted tools is changing almost daily. One such list: metaschool.so/articles/ai-code-editor

We have observed that relying too heavily on AI can sometimes lead to surprising results, such as non-human-friendly code, or excessive nesting levels, but that LLMs perform quite well with modular code. It can however struggle when working with larger files that have high complexity. It seems likely to me that over time, engineers will change their practices on code and logic organization to ensure it can work well with AI-assisted tools.

---

An open question we often discuss is the best approach to integrating AI into our development process. Some engineers choose to lead with AI, allowing it to generate initial drafts and then apply their human insight to refine the code. Others start with the human writing an initial stab, and then have the AI assist in reviewing and refinement. We acknowledge that individuals will use different approaches, at least in the near term. Perhaps eventually, tried and true methods will arise and become standard.

Even with these powerful AI tools, we recognize that nothing can replace the depth of understanding gained from real experience. It’s important to highlight that while AI can help junior developers learn quickly and develop their skills, real expertise comes from analyzing different development approaches, understanding architectural decisions, and knowing the right questions to ask. At Devblock, we’re combining the power of AI with our team’s extensive experience to deliver innovative, high-quality, maintainable solutions for our clients.

Interested in an AI-powered engineering engagement? Give us a call and let’s discuss!