Read Client Stories
Alex Kugell
Key Takeaways
- Python 3.14, released in October 2025, made GIL-optional builds officially supported, with roughly 3x speedups on multi-threaded CPU work.
- The tooling got rebuilt in Rust, and uv replaced most of the packaging toolchain in under two years, now exceeding pip in CI usage on major projects. Ruff consolidated linting and formatting into one binary.
- AI integration drives most new Python work, which is less about training models and more about building applications around them.
- Python is slow in ways that usually do not matter, since the heavy numerical work runs in C or Rust underneath.
- Web development has shifted toward FastAPI for new API work, while Django holds its position for full applications.
- Financial services use Python heavily for risk, pricing, and fraud work, and considerably less for the transaction path itself.
Python ranks as the most used programming language in most major developer surveys, which tells you very little about what people actually build with it.
But it definitely isn’t right for every instance. Knowing where Python is being used right now can help you decide where to use it in your own applications, and also help you in your search for a developer.
This article covers twelve application areas, the libraries behind each, and which companies run them in production.
We’ll also look at two changes from the past eighteen months that altered what Python can reasonably be used for.
At Trio, we specialize in fintech development, and our senior Python developers are experts in the language’s applications in financial technology.
We can provide insights and connect you with the right people for your requirements from our pre-vetted pool.
What is Python?
Guido van Rossum released Python in 1991 as an interpreted, high-level, general-purpose language built around readability. The design goal was something simple enough for beginners and capable enough for serious work, which is an unusual combination to actually achieve.
Three properties explain most of its spread:
Readability: Python code reads close to English, which lowers the cost of someone else maintaining it. When your team is made up of data scientists, backend engineers, and analysts who all touch the same repository, that matters more than elegance.
Libraries for nearly everything: The standard library covers a lot, and PyPI covers the rest. Work that would take a thousand lines elsewhere frequently takes ten.
Integration: Python works comfortably alongside C, C++, Java, and Rust, which is why performance-critical pieces can be written elsewhere and called from Python without friction. NumPy is a C library with a Python interface. Polars is a Rust one.
However, with these properties came some real trade-offs.
Python is slow at raw computation compared with compiled languages, uses more memory, and until recently could not use multiple CPU cores within a single process.
However, in reality we have found that the first two rarely matter because the heavy lifting happens in compiled extensions. The third also just changed.
What free-threaded Python changed.
For three decades, the Global Interpreter Lock prevented Python from executing bytecode on more than one core at a time within a process.
On a 32-core server, a single Python process used one core for computation, and the workaround was multiprocessing with its memory overhead and serialization costs.
Python 3.13 shipped an experimental free-threaded build. Python 3.14, released 7 October 2025, made it officially supported under PEP 779, with the single-threaded performance penalty down to roughly 5 to 10%.
Reported speedups on multi-threaded CPU-bound work average around 3x and scale close to linearly on genuinely parallel problems.
But you need to keep in mind that there are a couple of really important caveats.
The GIL is removable, not removed: standard builds still have it, and you opt in by running the separate free-threaded build, usually python3.14t.
On top of that, C-extension compatibility is still a massive blocker, since importing an extension without free-threading support silently re-enables the GIL, which means you can believe you are running free-threaded and not be.
Making free-threading the default has no PEP and no timeline. Realistic estimates land around 2028 or 2029.
Python 3.14 also shipped an experimental JIT compiler, which we really think is worth watching.
The tooling rebuilt itself in Rust
The second significant shift happened outside the language.
uv, Astral’s Rust-based package manager, replaced pip, pip-tools, virtualenv, pyenv and poetry for a large share of the ecosystem in under two years, running 10 to 100 times faster.
It has overtaken pip in CI usage on major projects. Ruff, from the same team, consolidated linting and formatting into a single binary fast enough to run on every keystroke.
OpenAI acquired Astral in March 2026, which raises reasonable governance questions about two tools that became infrastructure very quickly.
If your team is still on pip and a requirements.txt, uv is usually the cheapest meaningful improvement available, since CI time drops immediately and nothing about the code needs to change.
12 examples of Python applications
1. Web development
Python handles the server side through frameworks that cover different needs.
Django takes a batteries-included approach with an ORM, admin interface, authentication, and templating built in. It suits full applications where you want conventions rather than decisions. Instagram, Pinterest, Mozilla, National Geographic, and NASA all run Django somewhere.
FastAPI has become the default for new API work. It builds on type hints and Pydantic to generate validation and OpenAPI documentation automatically, and its async support suits services making many concurrent calls. For an API-only backend in 2026, this is the usual starting point.
Flask is still a capable micro-framework for small services, prototypes, and model demos, where Django’s structure would be overhead.
Just keep in mind that Python runs on the server, not in the browser. Client-side code is JavaScript, and the two interleave rather than compete.
2. Data science and analytics
The area where Python displaced established tools most completely.
pandas handles tabular manipulation, NumPy provides the array computing underneath nearly everything numerical, and Matplotlib, seaborn, and Plotly handle visualization.
Polars, written in Rust, has gained ground as a faster alternative to pandas for larger datasets, and DuckDB fits analytical queries over local files well.
Most of what makes this fast is not Python.
NumPy operations execute in compiled C, so a vectorized operation over a million rows runs at C speed from a Python call.
3. Machine learning and AI
PyTorch dominates research and increasingly production. TensorFlow retains an installed base and strong deployment tooling. scikit-learn covers classical machine learning, which still solves a large share of real problems that do not need deep learning.
The newer layer sits above model training. LangChain, LlamaIndex, and the OpenAI and Anthropic SDKs handle building applications around models: retrieval, prompt orchestration, tool use, and agent patterns.
This is where we have seen most of the new Python work in AI actually happens, since comparatively few teams train models from scratch and a great many integrate them.
4. Automation and scripting
The use case that quietly justifies Python inside most organizations.
Selenium and Playwright drive browsers, Requests and httpx handle HTTP, openpyxl manipulates spreadsheets, and the standard library covers file operations and scheduling.
DevOps engineers, analysts, and marketers all use Python to remove repetitive work, frequently without calling themselves programmers.
A finance team reconciling spreadsheet exports by hand each month is a near-universal pattern, and a hundred-line Python script usually ends it.
5. Game development
Python’s role here is mostly in scripting rather than engines. Civilization IV, Battlefield 2, The Sims 4, EVE Online, and World of Tanks all use Python for game logic and modding, with the performance-critical engine written in C++.
Pygame suits learning and small 2D projects. Serious commercial games generally reach for Unity or Unreal, with Python appearing in tooling and pipelines around them.
6. Desktop applications
Not Python’s strongest area, though the toolkits are capable. PyQt and PySide wrap Qt and produce genuinely professional interfaces.
Tkinter ships with Python and suits simple utilities. Kivy targets touch interfaces.
Packaging and distribution remain awkward compared with native development, which is why Python desktop applications are more common as internal tools than as consumer products.
7. Web scraping and data extraction
Beautiful Soup parses HTML, Scrapy handles larger crawling operations with concurrency and pipelines built in, and Playwright covers pages that need JavaScript execution.
Common applications include competitive price monitoring, aggregating job or property listings, gathering research data, and feeding content into retrieval systems for AI applications, which has become a substantial driver of scraping work.
Legal and ethical boundaries matter here, and vary by jurisdiction and by what the site’s terms say.
8. Cybersecurity and network tools
Scapy manipulates packets, Impacket handles network protocols, and the standard library’s socket module covers the basics.
Python’s speed of iteration suits security work well, since the task is usually writing a specific tool quickly rather than writing fast code.
Applications include penetration testing, malware analysis, log analysis, and automating security checks in a pipeline. Most published security tooling has a Python implementation somewhere.
9. IoT and embedded systems
MicroPython and CircuitPython run on microcontrollers, and full Python runs comfortably on a Raspberry Pi.
This suits prototyping, home automation, environmental sensors, agricultural monitoring, and industrial data collection.
Where power budgets or real-time guarantees are tight, we have found that C still wins. Where iteration speed matters more, Python is usually the best option.
10. Finance and fintech
Python is deeply embedded in financial services, though not uniformly across the stack.
Quantitative analysis and risk modeling rely on NumPy, pandas, SciPy, and statsmodels for pricing, portfolio optimization, value-at-risk calculations, and stress testing.
Fraud detection uses scikit-learn and PyTorch for the models, frequently with Python handling feature engineering while the scoring itself runs somewhere faster.
Algorithmic trading uses Python for research and backtesting, with execution often moving to C++ or Rust where microseconds count.
Regulatory reporting and reconciliation lean on Python’s data manipulation.
JPMorgan Chase built Athena on Python, Bank of America has a comparable Quartz platform, and Citadel, Two Sigma, and most quantitative funds use it heavily.
What Python generally does not do in finance is sit in the synchronous transaction path.
Payment authorization, ledger writes, and core banking tend to run on Java, Go, or C++, where predictable latency under load matters more than development speed.
Python sits beside that path, involved in scoring the risk, analysing the outcome, and even generating the report.
11. Business and enterprise applications
Internal tooling, ERP integration, reporting pipelines and data infrastructure.
Apache Airflow, itself written in Python, orchestrates a large share of enterprise data pipelines.
Streamlit and Dash let analysts build interactive dashboards without front-end work, which has quietly replaced a lot of internal tool development.
For example, Spotify uses Python across data infrastructure and recommendations, while Uber uses it in pricing and forecasting. Netflix also uses it extensively in data engineering and operations tooling.
12. Cloud and API integration
Python suits cloud automation well.
boto3 covers AWS, with equivalents for Azure and Google Cloud. Pulumi and the Terraform CDK allow infrastructure as code in Python rather than a separate configuration language.
Serverless functions in Python work across all major providers, and the startup penalty that once made this awkward has narrowed.
We have also seen Python remain the common choice for the glue code connecting internal systems to third-party APIs.
What Python is not good for
A balanced view helps, and most articles on this subject skip it.
Mobile applications: Kivy and BeeWare exist, and neither is a serious competitor to Swift, Kotlin, or React Native for production apps.
Browser-side code: PyScript and Pyodide can run Python in the browser via WebAssembly, which is genuinely interesting and not yet a reason to stop using JavaScript.
Hard real-time systems: Garbage collection introduces unpredictable pauses, which rules Python out where timing guarantees are absolute.
Latency-critical transaction paths: Covered above. Where a p99 budget is measured in single-digit milliseconds, compiled languages win.
Large codebases without type hints: Dynamic typing accelerates early development and costs you later. Type hints checked with mypy or pyright recover much of that, and teams that skip them on a growing codebase usually regret it.
Hiring Python developers
Someone who builds data pipelines, someone who trains models, someone who writes Django applications, and someone who automates infrastructure all answer to the title and have limited overlap in practice.
Worth being specific about which you need.
For financial work, the distinction between analytics experience and money-movement experience matters more than general Python fluency, and the two pools are different sizes.
We place senior Python developers with production experience in regulated financial environments, through staff augmentation or as a dedicated team.
If you are scoping a project or scaling a team, request a consult.
