Building with Claude Code, Feb '26

As a Python user and proud owner of ~3 dozen coding books, I was eager to test Claude Code with the media attention and hyperbolic buzz surrounding it. So, I built a web application with Claude Code (a strat-o-matic-like MLB Roster Builder for baseball fans), that purposefully incorporated frontend components outside my comfort zone. Here is my experience:

• Setup: Claude Code is an AI Agent working in your local terminal, so the setup is clunkier than button-pushing or prompting. But if you’re familiar with a command line interface, you’ll be running in 20 minutes.

• Initial reaction: Remarkably more powerful and agentic than coding alongside GenAI. It's like going from an aux cord to Apple CarPlay.

• Efficiency: This is the key metric - the project (a mobile-friendly, web-based, button-heavy React app) was chosen because it would’ve taken 3-5 weekends of development without AI. With Claude Code, launched in ~12 hours.

• Effectiveness: In my opinion, its effectiveness is largely determined by the user’s technical knowledge (including but not only coding ability). A user with an understanding of frontend, backend, designing a database, repositories, git commands, using API keys, reading log files, handling a syntax error, etc, can go further and more complex with Claude Code than a nontechnical user. This doesn’t mean a nontechnical operator cannot bring their idea to life, but Ethan Mollick would label developing with Claude Code a “cyborg activity”. It’s a mix of human and AI input for seamless, continuous progress. Thus, human input and technical skills are still quite important. (additional note: during development, I caught 2 calculation/math mismatches between the logic Claude Code summarized and the actual Python files it created)

• Individual Impact: The tool allows individuals to build, showcase, or share their domain expertise/ideas! It’s like an amplifier, that removes the technical hurdles for the user by subsidizing parts of software development they may lack. For example, I was able to focus on implementing a regression model to predict MLB Team Wins based on roster features, rather than getting caught up making complicated dropdown buttons with React on the app’s user interface.

• Business Impact: My initial 250-word prompt to Claude Code explained the goal and features of the intended web app. It responded with a plan: a summary of the files, along with the frontend, backend, and database software tools it would use. I could've accepted its plan by simply typing "1", and it would start cooking. But as a database expert, I noticed I didn't want to use Supabase. Also, what's Netlify? I want a simple frontend with Cloudflare Pages. I quickly concluded, Claude Code (and other Agentic Coding Tools) will make important vendor decisions for us. It will decide to use Google's database over the Amazon competitor, or vice versa. How Agentic Coding Tools make these decisions and how vendors may influence them, is a whole new form of SEO.

If you want to test the app, go for it: https://mlbrosterbuilder.com/. Use the Share button to download or send your roster, and click the Palette button to update your color scheme. Regading the baseball-specific nuances:

• MLB Roster Builder allows a baseball fan to design their ideal lineup and pitching rotation for an MLB team, choosing from the players on that team's 40-man roster and top prospects in their minor league system.

• Each player's statistics are 2026 projected stats from Fan Graphs. The Fan Graphs data is obtained via Python with pybaseball.

• After a full roster is filled on the app, a Regression Model predicts the number of Team Wins based on roster features such as total WAR, ERA, HR, AVR, FRV, and some others. This model is trained on the previous 10 seasons of MLB data, without any shortened seasons.

• For position players, lineup order and fielding position matter! (better lineup protection, more plate appearances, and experience playing a position boost player stats and impact Team Wins)