Python.

Where Python fits on a long engagement.

Python earns its place through libraries rather than language features. Pandas and NumPy for analysis, PyTorch and scikit-learn for modelling, Airflow for orchestration, FastAPI and Django for services. A team that already writes Python for one of those can usually pick up the next without a hiring round, which matters more on a three-year roadmap than on a project.

Its weak points are predictable and worth planning around. Raw request throughput is lower than a compiled runtime, packaging and environment drift cause more production incidents than the language itself does, and a large untyped codebase gets expensive to change. Type hints and a strict CI gate are usually the cheapest corrective.

What an assigned team does with Python.

On an assigned team Python is rarely one person's job. A backend engineer writes services in it, a data engineer writes pipelines in it, and an analyst writes notebooks in it, which means conventions have to be shared across disciplines rather than owned by one. Getting that agreed in the first month is cheaper than reconciling three styles in year two.

What changes over a long engagement is the proportion. Most roadmaps start with one Python discipline and add a second within the year, which is why the capacity is normally set up as a dedicated software development team covering several rather than as separate agreements per specialism.

What we use Python for.

• An API and its data pipeline in one language A backend team and a data team sharing conventions, review standards and utility code instead of maintaining two ecosystems.
• Model training that reaches production The same language from notebook to served endpoint, which removes the rewrite step where most machine learning work stalls.
• Replacing spreadsheet-driven process Recurring manual reporting rebuilt as scheduled jobs with error handling, so a finance close stops depending on one person being available.

How Python capacity is assigned.

Python capacity is assigned inside whichever discipline owns the work rather than as a language specialism. Where the constraint is pipelines and warehouse models, that is managed data services; where it is model-backed features, AI engineering services.