Unpopular opinion: you don’t need Python to build powerful AI features anymore.
Unless you’re training custom models (and let’s be honest, most product teams aren’t), Python is no longer the obvious choice.
At Nimble Studio, we help companies in healthcare, retail, energy and beyond build and ship AI-enabled digital experiences. And nearly all of those features are powered by TypeScript, not Python.
Here’s why.
Python Had Its Moment
In the early days of large language models, Python was the natural choice. Most breakthroughs came from machine learning researchers working in tools like PyTorch or TensorFlow. Python was their playground and for that phase of innovation, it made sense.
But we’ve shifted to a new phase.
Today, most AI features inside products don’t involve model training. They’re built by calling APIs like OpenAI, Anthropic, or Mistral. The work has moved from the research lab to the product team. From tuning parameters to designing user flows. From Python to JavaScript.
Many teams start with tools like LangChain to build basic AI features. But once you move to agents and more complex logic, tools like LangGraph (and its JS versions) become more relevant. That shift is where TypeScript developers shine.
A Stack Built for Shipping AI.
When you’re building AI into web products, you need speed, flexibility, and stability. You need tools that plug easily into your frontend and backend. You need dev-friendly workflows.
That’s why we started building with LangGraph.js, a JavaScript port of the Python-native LangGraph. It was a good start, but over time, we hit its limits.
We’ve since moved to Mastra, and haven’t looked back.
Here’s a quick comparison:
LangGraph.js (where we started)
- A direct port from Python, with Python-style thinking
- Frequent breaking changes with major upgrades
- No client-side SDK
- Outdated docs and complex setup
- Steep learning curve, especially for frontend developers
Mastra (what we use now)
- Built natively for TypeScript developers
- Plug-and-play architecture, flexible for custom logic
- Comes with an MCP server that integrates directly into Cursor (our preferred code editor)
- Clear documentation and fast setup
- High abstraction with a low barrier to entry
What Mastra Lets Us Do
Mastra gives us everything we need to prototype, ship, and scale AI-powered features inside modern web products. It comes with built-in workflows, agent logic, RAG support, tracing, retries, and human-in-the-loop fallbacks.
It also plays well with providers like OpenAI, Anthropic, and Gemini making it easy to test and switch models depending on the use case.
And the best part? Our team doesn’t need to context-switch out of our existing TypeScript stack. Mastra meets us where we already are.
How This Fits Nimble’s Way of Working
At Nimble, we don’t believe in long roadmaps or big reveals. We work in tight feedback loops with our clients, whether we’re building a healthcare assistant, smart search for ecommerce, or back-office automation for energy providers.
Mastra fits that approach perfectly. It lets us move from idea to working AI feature in a matter of days, not weeks. No model tuning. No waiting. Just fast, user-focused product work.
Final Thought
If you’re building AI features for real users (especially on the web) TypeScript is no longer just an option. It’s often the best tool for the job.
Python still has its place in research. But for shipping AI in modern products?
We’re all in on TypeScript.
And thanks to tools like Mastra, our team at Nimble is shipping faster, smarter, and with more confidence than ever.