Bard can now code and put that code in Colab for you.

Google’s Bard has gained a practical new capability: it can generate Python code and export that code directly into Google Colab, turning prompts into runnable notebooks. The update matters because it closes a major gap that frustrated early Bard users—public Bard couldn’t write code—while also aligning with Google’s broader challenge of serving AI models at scale using smaller, more efficient versions.

In a hands-on walkthrough, Bard is shown producing code quickly for straightforward tasks, such as writing a Python function to test whether a number is prime. The workflow is the key upgrade: after Bard drafts the code, a user can click “export to Colab,” which saves a Colab notebook to Google Drive. Opening the notebook reveals the generated code ready to run, and the example demonstrates the function working inside the Colab environment.

The same coding workflow scales to more complex, multi-step projects. Bard can generate Colab cells that create a SQLite database, define a table named “people,” and populate it with random data fields including name, age, gender, and favorite ice cream flavor. It can then produce SQL queries to compute useful statistics—average age, counts by gender, and the most popular ice cream flavor. Bard can also translate those SQL queries into Python so the results can be computed and displayed within the notebook.

A notable part of the process is debugging. When the exported Colab code fails due to a missing import (a NameError around a random integer function), Bard can be prompted again to fix the code. The corrected notebook is then rerun after removing duplicate database creation logic, and the analysis outputs appear as expected—showing that the system can iterate toward working code rather than stopping at a first draft.

Bard also supports a “print the table” style request, formatting the full SQLite table as a pandas DataFrame. The walkthrough highlights that the interface often includes text boxes alongside code cells, which can help learners understand what the notebook is doing.

Beyond code generation, Bard can work with GitHub repositories: it can summarize what a repo does, connect it to related papers, and provide explanations of underlying concepts. An example uses a repository tied to “colT5 attention,” where Bard links the work to a paper and explains the motivation and approach—while also warning that hallucinations are possible and should be treated cautiously.

When asked to implement “flash attention” in PyTorch, Bard provides code but the implementation is flagged as incorrect, apparently mixing ideas from a different paper. Still, the overall pattern is clear: Bard can generate starting points, cite sources when available, and help users explore topics like transformers attention mechanisms and even define terms such as LangChain.

Access is rolling out, with Google working to serve the model in more countries and gradually expand availability. For users who already have access, the practical takeaway is immediate: prompts can now become runnable Python notebooks in Colab, with iterative fixes and repository/paper explanations layered on top.