Gemini — understanding through editing

Google · 2024

People wanted to check the data behind a generated chart, change its labels, or try another chart type. My work on Gemini covered how people could interact with generated graphs and code, within the limits of what the model could do.

The question was how editing could become part of understanding the result.

Make the result editable

People were asking for a different chart type, updated labels, or a change of color. These small refinements could interrupt the workflow and leave the result feeling unfinished.

We developed tap-and-click editing around those common needs. The version that shipped supported a limited set of changes, including chart type and labels.

The choices needed care, too. Chart types and color families could leave people unsure what to choose. The direction was a smaller set of options, with room to try another view and understand what it showed.

Keep the source within reach

A chart needed a way back to its data. People needed to check how rows had been parsed, move between the chart and table, and export when necessary.

Source access gave people something to inspect alongside the generated result. The fuller graph experience, combining editing and source context, had a limited release. Asking the model about the asset was part of the broader vision and did not ship.

The same question in code

Code was the other part of my work. The shipped version supported sharing and export; a limited release added editing, running code, logs, and explanation. Co-development remained an unshipped direction.

Both experiences were about understanding through working with a result. One participant, using graphs, put it this way: