An open-source interactive computing environment that combines executable code, rich text, visualizations, and data in a single document, making it the standard tool for data science exploration, AI model development, and analytical reporting in agency technical workflows. Jupyter Notebooks are where most agency data science and AI development work lives before it moves to production.
Also known as notebook, Jupyter, .ipynb
A Jupyter Notebook organizes work into cells, each of which can contain either executable code or markdown text with formatting, equations, and embedded media. Code cells are executed individually or sequentially, with outputs including plots, tables, and text rendered inline directly below the code that produced them. This interactivity makes Jupyter the standard environment for data exploration: a data scientist can load a dataset, run a summary statistics cell, inspect the output, run a visualization cell, iterate on it, and document their observations in a text cell, all in a single document that tells the story of the analysis alongside its code and results.
Jupyter supports multiple programming languages through the kernel system, with Python being by far the most common for AI and data science work due to the breadth of its machine learning ecosystem. R, Julia, and Scala kernels are also available. The ipynb file format that Jupyter uses is a JSON document storing each cell’s type, content, and most recent output, enabling notebooks to be version-controlled, shared, and re-executed. This shareability is one of Jupyter’s most valuable properties for agency work: a notebook that contains both the code for an analysis and its outputs serves as a self-documenting analytical artifact that communicates the methodology and results simultaneously to a technical audience.
Jupyter has evolved from a local tool to a cloud-hosted environment through platforms including Google Colab, which provides free access to GPU instances for model training; JupyterHub, which enables shared notebook environments for teams; and managed notebook services in AWS, Google Cloud, and Azure that integrate with cloud storage and compute. These cloud-hosted versions remove the installation and compute infrastructure management burden from individual practitioners, enabling agency data science teams to collaborate on notebooks and run computationally intensive work without managing their own hardware.
Data analysis, model development, and AI experimentation in agencies increasingly happens in notebook environments, and deliverables from data science work, including exploratory analyses, model validation reports, and performance deep-dives, are often first produced as notebooks. A working ad agency that is fluent with Jupyter Notebooks can produce analytical work that is transparent, reproducible, and directly communicable to technical clients and stakeholders, rather than producing opaque results from black-box tools.
Notebooks serve as living documentation for analytical decisions. A well-structured notebook documents not just the final analysis but the exploratory process: the data quality checks that were run, the feature engineering approaches that were tried and abandoned, the model variants that were tested, and the evaluation criteria that led to the final recommendation. This documentation is valuable for client deliveries, internal knowledge transfer, and regulatory audits that require a record of how an analytical conclusion was reached.
Shareable notebooks improve client communication of technical results. Rather than delivering a static slide deck with charts produced by an opaque process, delivering a Jupyter Notebook that clients or their technical teams can re-execute, inspect, and extend provides a higher level of transparency and enables the client to validate the analysis independently. For technically sophisticated clients who care about methodology, a notebook delivery demonstrates rigor in a way that a presentation cannot.
Google Colab reduces the barrier to GPU-accelerated model development for agency teams. Colab provides free access to GPU and TPU instances that enable running model training experiments that would be impractical on local hardware. For agencies that are experimenting with fine-tuning or evaluating AI models without dedicated cloud infrastructure, Colab provides a path to GPU-accelerated experimentation without capital expenditure or complex infrastructure setup, enabling more team members to participate in AI model development work.
An agency data scientist is conducting a quarterly analysis of a retail client’s customer lifetime value model performance. Rather than producing the analysis in a business intelligence tool that generates static charts, they build a Jupyter Notebook that loads the most recent customer data, recomputes model predictions, generates all validation metrics and visualizations, and outputs a structured commentary on model performance versus the prior quarter. The notebook is designed to be re-executable: a client team member with basic Python knowledge can re-run all cells to reproduce the analysis on updated data, verify the computations, or extend the analysis with additional queries. The agency delivers both a PDF export of the notebook for stakeholders who want the narrative and a link to the notebook file for the client’s technical analytics team. The client team uses the notebook as a template for their own monthly model monitoring process, extending it with company-specific business logic that they add in new cells, reducing their ongoing monitoring work from a multi-day manual process to a 30-minute notebook re-execution with review.
The generative AI foundations module covers the tools and practices of AI model development, including the notebook-based workflows that make analysis transparent, reproducible, and communicable to technical clients and stakeholders.