tldr;

It's incredible how fast AI has become part of our daily lives, both inside and outside our work as developers. I have been a pretty quick adopter of AI tools, and in many ways it's allowed me to learn more about different topics, solve problems I was facing for work, and build applications for personal as well as professional use that I just didn't have time to build before. What I’ve found, however, is that a well-curated set of tools is essential. It’s not just about using AI; it’s about understanding the specific role each tool plays in your workflow to ensure you actually cross the finish line. In this article, I'll share my process of how I've built several Angular and full stack applications so far this year using AI tools, and the specific tools that I use to accomplish different tasks.

For this article, I will use a library app as the demo project, one that keeps track of books that you own, have read, etc. It's not one I've built, so I'll use it as an example of the process I go through to build a new application.

First Step: A Good Plan

Every good project starts with a well-defined plan and clear goals and desired outcomes. This is obviously not unique to software development; it applies to every project we want to take on in our lives. Thus, I've found that just jumping into Antigravity or Claude and saying something along the lines of "build me an app that manages my library of books" will not get you very far. Or at least, it won't get you very far with a usable result. Sometimes the AI model will prompt you for some more information, but it also may just start to build something for you before getting all the details it needs. To prevent this from happening, I like to develop a good plan and project outline before anything starts.

To make this plan, I still will start with an AI model. I generally use Gemini for things like this, but any of the offerings out there should work. My initial prompt will be something like this:

I want to build an application that will allow me to manage my library of books. I want to be able to add books, edit them, and delete them. I should be able to rate them after reading them, tag them for organization, and other library-type actions. For the stack, I want an Nx workspace with a Nest API and an Angular frontend. I want to use PrimeNg for the component library and Tailwind for styling. Make sure to use modern Angular patterns, including control flow syntax and Signals. Ask me questions so we can build out a good plan for this project.

This prompt gives the AI model a good starting point. It's not making any decisions yet, and it's not guessing at what I want. From here, it can ask some follow up questions to make sure we include all the features that I want (i.e. "Would you like to keep track of loaning your books to friends?" or "Do you want to keep track of the format of the book?"). Many times, the model will help me think of features that I didn't think of, or it asks about something that I know I do not want to include. All of this gives the model a clear understanding of what I want to build. Even for a big project, the scope will be much more well defined after this process.

I will continue with this process until I feel like I have all the features I want included, and all the details ironed out. I will then ask the model to give me a technical requirements document based on the project we've designed. I'll take the output and save it as a markdown file on my computer. This document will serve as the reference point for the beginning of my project. Sometimes the project begins to veer off course, and I can get back on track by referring back to the TRD. Other times I allow it to veer off because we have come up with a better plan.

Second Step: The Right Tools

After generating the TRD, I'm ready to get started on building the project. You can do some of this on your own, or you can turn it all over to an AI model. However, you need to have a good tool in place for any amount of AI involvement. I primarily use three different tools for this step: Antigravity, Gemini CLI, and Claude. Each of them has its strengths and weaknesses, but I have pretty successfully used them together to build applications. Antigravity is great if you like to work in an IDE, where you're able to easily view the files that are being updated alongside a chat window with the agent. Gemini CLI and Claude are great if you are comfortable being in a terminal session and chatting with the agent while not paying as much attention to the file structure without opening the repo up in an IDE. Neither of these tools is "better" than the other, it merely depends on what you feel most comfortable with.

I should point out, I have a Gemini Pro subscription due to a purchase of a phone last year, and I paid for a Claude Pro license as well. If you don't have a subscription to a tool like this, your ability to use AI for development will be severely limited.

Once you have your selected tool installed and ready to go, you are ready to start building. This is where a couple more tools come into play. I use two tools pretty frequently to help with the actual building of the app. The first is GSD, or get-shit-done, by gsd-build. It's described as "a light-weight and powerful meta-prompting, context engineering and spec-driven development system". Essentially, it will take a set of requirements you give it and generate a detailed plan broken down into milestones and phases. During each phase, you will be asked several questions where you can give your answers or select recommended options from the model. The depth and detail that GSD goes into has been surprisingly good in my experience, and is another way that I have made sure to not forget necessary features along the way.

Another tool that I recently started using but has helped with the UI/UX side of things is the UI/UX Pro Max skill from next-level-builder. I've only used this a couple of times, but it has done a great job at setting up a reasonable style guide and normalizing the use of classes and other design tokens instead of the general HTML and CSS written by AI models. This is a new tool for me, but some coworkers of mine have been using it extensively and have been happy with the results.

Another tool that I was recently introduced to and comes highly recommended is called gstack by Garry Tan. The goal with gstack is to turn your AI model into a full team to help with planning a project. You get a CEO, designer, engineering manager, release engineer, and more all in one tool. Where GSD takes what you want and plans that out, without much push back or questioning your motivations, gstack is designed to put you through a real planning session: why are you building this, is there a market for it, etc. Like I said, I was just introduced to this tool, but it seems like it could be a great way to make sure you have a solid plan in place before you start building.

As an example of gstack, my one time using it so far, I told it I wanted to expound on a POC project I'd started for work. I wanted to build out the project more in depth, and it pushed back asking if there was real buy in for the project, or if I'd be wasting my time on something that would never land. It was actually a great example of what the tool can do if you're starting on a brand new project.

MCP Servers

I don't have a lot of specific MCP servers that I use, but look for ones that are designed for the tools you use. GitHub is a good one so it can create issues, close issues, push and pull, create PRs, etc. It's really helpful to have everything tracked in GitHub. We use ClickUp at work for tracking tickets, so I use that MCP server as well. I've heard good things about the Sentry MCP server as well, but haven't used it for my projects yet. Regardless, many tools that you are already using have MCP servers. Download and run them while you're working on the project and I'm sure your project will be better off for it.

Third Step: Building the App

After all the prep work you've done up to this point, you're ready to get started on the project. Because of that prep work you'll be set up to make quick progress on the project. However, I still don't recommend just letting the agent write code without any human intervention. For the most part, I do let it write all the code, and then as it stops to ask for reviews I'll ensure I do that. If it wants me to review a given implementation in the code, I'll do that. If it just needs a human to test the interaction in the application, I'll do that as well. This is important because many times it thinks it's implemented a feature correctly and I find that it has not. The longer you let it go without checking, the more work you will have in the future to get back on track.

I also like to run the linting, unit tests, and end to end tests frequently to ensure that we are not breaking things as we move forward. I'll run them manually after each GSD phase, but I'll also instruct the agent to run the tests as it goes along as well. This is one more way to ensure we have quality code that does what we expect it to do.

Conclusion

We live in a new world, one different than even just a couple years ago. The ability to increase our output with the use of AI agents is exciting. This year I've built multiple full stack applications and improved some open source libraries that I maintain using AI tools. If not for those tools, I wouldn't have finished any of those projects. It's easy to want to push back on using these tools, but I encourage everyone I talk to to embrace the change and to learn how to effectively use the tools in their day to day work. The more you use them, the more comfortable you will be and the more quality code you will be able to produce.

Referenced Resources & Tools

AI Development Platforms

• Antigravity – Google's agent-first IDE (fork of VS Code) optimized for Gemini and autonomous coding tasks.
• Claude – Anthropic’s LLM, highly effective for code generation and architectural reasoning.
• Gemini – Google’s multimodal AI, excellent for initial planning and large-context window tasks.

Planning & Workflow Skills

• GSD (get-shit-done) – A spec-driven development system for breaking down projects into actionable milestones and phases.
• gstack by Garry Tan – An open-source toolkit that adds "engineering team" personas (CEO, QA, etc.) to your AI workflow.
• UI/UX Pro Max Skill – An AI skill designed to provide high-level design intelligence and consistent styling tokens.

Model Context Protocol (MCP) Servers

• GitHub MCP – Standard server for managing issues, PRs, and repository actions.
• ClickUp MCP – Integrate your workspace tasks and docs directly into your AI chat context.
• Sentry MCP – Helpful for feeding error logs and performance issues directly to your AI agent for debugging.