AI-powered tools are transforming the way we code, and I recently got a chance to dive into this revolution with Codeium’s Windsurf IDE. My journey spanned two exciting projects: updating the theme of my mdsinabox.com project and building a Terraform provider for MotherDuck. Each project offered unique insights into the capabilities and limitations of AI-enhanced development. It should be noted that I did pay for the “Pro Plan” as you get rate limited really quickly on the free tier.
Project 1: Updating the Theme on mdsinabox.com
My first project involved updating the theme of my evidence.dev project. Evidence.dev is a Svelte-based app that integrates DuckDB and charting (via ECharts). Styling it involves navigating between CSS, Svelte, TypeScript, and SQL—a perfect storm of complexity that seemed tailor-made for Windsurf’s AI workflows.
I aimed to update the theme fonts to use serif fonts for certain elements and sans-serif fonts for others. Initially, I asked the editor to update these fonts, but it failed to detect that the font settings were managed through Tailwind CSS—a fact I didn’t know either at the time. We wasted considerable time searching for where to set the fonts.
After a frustrating period of trial and error of pouring over project internals, and later reading documentation, I realized that Tailwind CSS controlled the fonts. Once I instructed the editor about Tailwind, it identified the necessary changes immediately, and we were back on track.
However, one gripe remained: Windsurf’s model didn’t include the build files for the Evidence static site, so I had to manually copy files to another directory for it to work. Additionally, debugging errors using the browser’s source view proved more efficient than relying on the editor. These limitations were a bit frustrating, but the experience highlighted the importance of understanding your project’s architecture and guiding AI tools appropriately. Access to a browser emulator would massively improve the debugging experience.
Project 2: Building a Terraform Provider for MotherDuck
The second project was sparked by a potential customer’s request for a Terraform provider for MotherDuck. While I was familiar with Terraform conceptually, I’d never used it before. With the recent launch of our REST API at MotherDuck, this felt like the perfect opportunity to explore its capabilities.
I instructed Windsurf, “I want to make a Terraform provider. Use the API docs at this URL to create it.” The editor sprang into action, setting up the environment and framing the provider. While its initial implementation of the REST API was overly generic and didn’t work, the tool’s ability to see the entire codebase end-to-end made it relatively straightforward to refine. I did have to interject and say “here is an example curl request that I know works, make it work like this” which was enough to get it unstuck.
As an aside, observing it at times was quite comical as it seemed to take approaches that were obvious incorrect, especially when I was dealing with some invalid authorization tokens. It would almost say “well I trust that my handler has given me a valid token, so it must be something else” and just start doing things that were obviously not going to work.
Anyway, once the main Terraform file was built, I tasked the editor with writing tests to validate its functionality. It recommended Go, a language I had no prior experience with, and even set up the environment for it. Through a mix of trial and error and manual intervention (particularly to address SQL syntax issues like the invalid ‘attach if not exists’ statement in MotherDuck), I managed to get everything working. From start to finish, including testing, the entire process took around four hours—which seemed pretty decent given my experience level.
Conclusion
My experience with Codeium’s Windsurf IDE revealed both the promise and the current limitations of AI-enhanced development. The ability to seamlessly navigate between languages and frameworks, quickly scaffold projects, and even tackle unfamiliar domains like Go was incredibly empowering. However, there were moments of friction—misunderstandings about project architecture, limitations in accessing build files, and occasional struggles with syntax. Getting these models into the right context quickly is pretty difficult with projects that have lots of dependencies and overall my projects are fairly low complexity.
Still, it’s remarkable how far we’ve come. AI-enabled editors like Windsurf are not just tools but collaborative partners, accelerating development and enabling us to take on challenges that might have otherwise seemed impossible. As these technologies continue to mature, I can’t wait to see how I can use them to build even more fun projects.