Images by the author | chatgpt
introduction
For years, Google Colab has existed as the basis for data scientists, machine learning engineers, students and researchers. It democratizes its most important computing resources in today's world, such as Graphic Processing Units (GPUs) and Tensor Processing Units (TPUs), and offers a free, no-configured host Jupyter notebook environment in your browser. The platform contributes to everything from learning Python and Tensorflow to developing and training modern neural networks. However, the artificial intelligence landscape is evolving at an incredible pace, and the tools we use must evolve with it.
Recognizing this shift, Google has announced its rethinking AI-First Colab. Announced at Google I/O 2025 and now accessible, this new iteration moves beyond the simple, hosted coding environment for becoming an AI-powered development workflow partner. By integrating the power of Gemini, Colab can now act as an agent collaborator who can understand code, intent and goals, lowering the barriers to tackling today's data problems. This is more than just an update. It is a fundamental change in the way data science and machine learning development can be approached.
Take a closer look at Google Colab's new AI features and find ways to use them to increase productivity in your daily data workflow.
Why AI-First is a game changer
Traditional machine learning workflows can be laborious. This includes a set of clear and repetitive tasks, including exploratory data analysis, data cleaning and preparation, functional engineering, algorithm selection, hyperparameter tuning, model training, model evaluation, and more. Each step requires not only deep domain knowledge, but also important time investments in written code, consulting documentation and debugging.
The new AI-first environment, like Colab, aims to significantly compress this workflow and embed AI in the development environment itself. Early use of these new AI-powered features suggests a double increase in user efficiency, transforming them into manual labor time-guided conversational experiences, allowing them to focus on more creative and important aspects of their work.
Consider these common development hurdles.
- Repeated coding: Creating code to load data, cleaning up missing values, or generating standard plots is a necessary but boring part of the process
- “Blank Page” Issues: Staring at empty notes and trying to find the best libraries and features for a particular task can be daunting, especially for newcomers.
- Debug Hell: Unclear Error Messages can derail your progress for hours when searching for forums and documents and searching solutions
- Complex Visualization: Creating quality charts for publications often requires extensive adjustments to plot library parameters. This is a task that diverts from actual data exploration
The new AI-First Colab acts as a pair programmer that can directly address these issues, suggest code generation, modifications, and help automate the entire analytical workflow. This paradigm shift means that time spent on coding mechanisms takes less time, and strategic thinking, hypothesis testing, and interpreting results.
Colab's Core AI Features
With Gemini 2.5 Flash currently running, we'll introduce three specific AI features that Colab offers to make your workflow easier.
1. Iterative Queries and Intelligent Support
At the heart of the new experience is the Gemini chat interface. You can find it via the Gemini Spark icon in the bottom toolbar for a quick prompt, or in the side panel for a detailed discussion. This is more than just a simple chatbot. It is context aware and can perform a variety of tasks, such as:
- Code Generation from Natural Language: Just explain what you want to do and Colab will generate the code you need. This can range from simple functions to refactoring the entire note. This feature significantly reduces the time spent writing boilerplate and iterative code.
- Explore the Library: Should I use a new library? Ask Colab for instructions and sample use based on the current notebook context.
- Intelligent Error Fix: If an error occurs, Colab not only identifies it, but also suggests a fix, presents the proposed code changes in a clear DIFF view, and can see and accept the changes.
2. Next-Generation Data Science Agent
The upgraded Data Science Agent (DSA) is another welcome addition to Colab. DSAs can autonomously perform complex analytical tasks from start to finish. Simply ask and you can trigger the complete workflow. The agents are:
- Generate a plan: An overview of the steps required to achieve your goals
- Run code: Write and execute the required Python code in multiple cells
- Reason for the results: Analyze the output to inform you of the next step
- Current survey results: Summary of the survey results and present them to you
DSA allows for interactive feedback during execution, refine or reroute processes so that analysis aligns with goals throughout the process. This creates complex tasks, such as shooting raw datasets and performing conversational processes with streamlined end-to-end cleaning, feature analysis, model training, and evaluation.
3. Transforming and visualization
It's easy to refactor or modify existing code. Please explain the changes you need in natural language. Colab identifies the associated code blocks and proposes necessary changes to the DIFF view for approval.
Furthermore, visualizing data, an important but often boring part of data exploration, is now easy. Users can ask Colab to graph the data, and the agent generates clearly labeled charts without the need to manually tackle the finer points of libraries such as Matplotlib and Seaborn.
Start a new AI-First Colab
Google has simply accessed the new Colab features. There are no complicated setups or waitlists. They're just there and ready to use – even in the free tier.
Once you've logged in to Colab with your notebook open, look for the Gemini Spark icon in the bottom toolbar. Click the icon to interact with the notebook, create a coding request, or ask a question.
Although these features are available in the free tier, it is important to note that the paid tier offers more reliable access, longer rantime, and faster GPUs.
Walkthrough with the new AI-First Colab
Let's test your DSA using some of the Colab sample data.
To that end, I opened the Gemini chat panel and entered the following simple exploratory data analysis task prompt (I don't need to ask Colab to use the DSA feature, but please note that I did anyway):
Let's analyze some data! Use the Data Science Agent feature to examine and report sample data in the Sample_data folder (/content/sample_data/anscombe.json).
New Google Colab UI with Gemini Chat Panel displayed (click to enlarge)
Colab's data science agent responded with the following plan:
- Load the dataset: Load the anscombe.json file into a Pandas dataframe.
- Explore a dataset: View the first few rows, check the data type, get an overview of the dataset, and understand its structure and content.
- Calculate summary statistics: Calculate descriptive statistics (mean, median, standard deviation, etc.) for each of the four datasets in the ANSCombe.json file.
- Dataset Visualization: Create a scatter plot for each of the four datasets to visualize the relationship between X and Y variables.
- Report your findings: Summary important findings from analysis and visualization, highlighting the importance of visualization in addition to summary statistics.
- Finishing Task: Place your findings into a format that anyone can read from previous stages.
The agent started executing the cell's code for each cell. If you run into a date format that you're not sure, you might pause and ask for clarification. Provides visualization – a combination task that can involve a significant amount of manual coding and debugging.
Final thoughts
Reimagined Colab marks the milestones of Google's journey towards more intuitive and powerful development tools, particularly the field of data science. By incorporating agent collaborators at the heart of Colab Notebook Experience, Google accelerated the work of experts and created a platform that makes the world of data science and machine learning more accessible to anyone. It may not be the authentic atmosphere coding that WW knows in other settings, but Colab offers something called vibe analysis…or Vibe Notebooking?
The future of coding is collab and with Colab, AI partners are now clicking and leaving the prompt.
Matthew Mayo (@mattmayo13) Get a Master's degree in Computer Science and a Graduate Diploma in Data Mining. As editor-in-chief of Kdnuggets & Statology and contributor to Machine Learning Mastery, Matthew aims to provide access to complex concepts of data science. His professional interests include exploring natural language processing, language models, machine learning algorithms, and emerging AI. He is driven by his mission to democratize the knowledge of the data science community. Matthew has been coding since he was six years old.
