At iGovTT, we are always learning, and part of that learning experience is to seek out tools that can benefit not only iGovTT and its services but also the government of Trinidad and Tobago as a whole. At iGovTT, we see AI as a tool that can exponentially increase productivity and efficiency. A direct result of proper AI usage is better products delivered in a shorter time frame, and persons interacting with government services are satisfied with our service delivery.
The Software Development Unit in iGovTT has developed an AI strategy that follows uses conventional policy of data usage within Trinidad and Tobago aka the Data protection act as a guideline. Using this as our north star for data usage, we can embark on leveraging this tool to show its value, specifically in software development.
As you all may know, iGovTT has a somewhat new development unit. Even though we have been developing solutions over the past three to four years or even more, there’s now a new development arm specifically created to develop solutions for the government of Trinidad and Tobago.
Thus, iGovTT embarked on a quest to find the ideal AI assistant for our development team, to have AI be our 24/7 code reviewer. This blog post will walk you through our exploration of various AI tools and share insights from our hands-on experiences.
The Criteria
The AI solution needs to satisfy the following needs:
- AI suggests code (predictions)
- Code review for errors, security and efficiency
- Generate code to help enhance the features of an application.
- Satisfies data usage policy (no data to be used for LLM training)
The Contenders
We focused our evaluation on three primary AI-powered development tools:
- Cursor AI
- Codeium (VS Code Extension)
- ChatGPT
- Blackbox AI
Let’s dive into these tools and see how they stack up.
Cursor AI
Cursor AI is a standalone IDE built from the ground up with AI integration in mind. It’s essentially a fork of VS Code, offering a familiar environment for many developers.
Pros:
- Excellent code generation and chat functionality
- Highly accurate code predictions (estimated 90% accuracy)
- Strong context awareness across multiple files
- Seamless integration of AI features into the development workflow
- Compatible with VS Code extensions
Cons:
- Requires switching from your current IDE, which may not be ideal for all developers
- Cost (subscription costs are not ideal for government entities)
Our developers were particularly impressed with Cursor AI’s prediction capabilities and context awareness. One developer (shout out to Dev) noted, “Overall, Cursor AI has been the most time-saving with its predictions and context awareness. The IDE is also close enough to VS Code that there was no learning curve to start using it.”
The tool’s ability to understand context across multiple files was a standout feature. One team member said, “It’s very good at context awareness when writing related code in different files.”
Cursor AI also showed its strength in complex tasks. A developer shared this experience: “I asked it to add limitations to specific feature. It added IDs to the related fields, created JavaScript to do the checks, and even added a popup message that would be activated if the criteria for the feature were met.”
The iterative nature of working with Cursor AI was also appreciated: “It shows you a preview of the code changes and lets you refine your request too if needed. You can keep adding to a request to give it more guidance and get exactly what you want.”
Our team was thoroughly impressed with code quality: “10/10 very clean code. Couldn’t find fault with generated code.”
Codeium (VS Code Extension)
Codeium is an AI-powered extension for VS Code, offering code completion and generation capabilities without the need to switch IDEs.
Pros:
- Integrates directly into VS Code
- Decent code predictions within a single file
- No need to switch development environments
Cons:
- Limited context awareness across multiple files
- Chat feature requires consent to use code for training
Our experience with Codeium was mixed. While it provided valuable functionality, it didn’t quite match up to our criteria. As one developer noted, “Used this previously, the limitation was prediction only. I did not use the chat feature as that depended on giving consent to use the code for training. The prediction was very decent, but it lacks the context awareness.”
The tool’s strength lies in its predictions within a single file, but it falls short when working across multiple files.
ChatGPT Chat
While not a dedicated coding tool, ChatGPT has become a popular resource for developers seeking general coding advice and ideas.
Pros:
- Excellent for generating example code and brainstorming ideas
- Versatile and can handle a wide range of programming-related queries
Cons:
- Lacks direct integration with development environments
- It will require more manual effort to implement suggestions
Our team found ChatGPT to be a valuable complementary tool, particularly for ideation and problem-solving. One developer stated, “It is good for general ideas and generating example code that can be built upon. I use it to bounce ideas off and get alternative suggestions based on what I’m trying to do.”
Additional Considerations
During our exploration, we also looked into Blackbox AI, another promising tool in the AI-assisted development space. While we didn’t conduct an in-depth evaluation at the time of writing, it’s worth mentioning its key features:
- Code Generation & Completion: Offers automatic code generation based on context and real-time code suggestions.
- Code Chat: Allows interactive conversations for code generation, bug fixing, and refactoring.
- Additional Utilities: Includes features like code search, commit message generation, and code-related image analysis.
Blackbox AI seems very similar to Cursor AI. Blackbox AI is particularly suited for speeding up development tasks such as fixing bugs, generating code, and refactoring, making it a potential asset in continuous integration workflows. Regarding transparency in how each application uses your data, Cursor AI better meets our criteria. Cursor AI explicitly states on its website that only a tiny amount of your data (code) is sent to a language model. It sends no more than 300 lines, which is just a small portion of your enterprise application. It is also SoC 2 compliant and GDPR compliant.
Conclusion
We are still experimenting and learning; this is a healthy place to be when using AI within the workspace. Giving guardrails, seeing what works on a small scale, documenting the learnings (including new guardrails), and gradually scaling up from prompting on ChatGPT to using native tools that bring AI into your everyday workflow and then expand what we have found to build better products/workflows and thus become more productive, more effective and provide more value to the citizens of Trinidad and Tobago.
However, we can’t end this article without stating that it’s important to note that the effectiveness of these tools can vary depending on your specific development needs and workflow. We recommend trying different options to find the best fit for your team.
As AI tools evolve, we’re excited about their potential to enhance our development processes and boost productivity. Have you tried any of these tools?
