Google admits biggest Gemini app problems, promises major fixes
Deep Research users asked for the ability to export reports to NotebookLM and switch models mid-chat
Google's Josh Woodward, the VP overseeing the Gemini app, did something most tech executives avoid: he publicly asked users to list everything wrong with his product.
On July 8, he posted a blunt question on X, asking what people were surprised Gemini still couldn't do well and what Google should have fixed long ago.
Over 1,400 responses came flooding in within 12 hours, along with more than 300 more by the next morning, resulting in a tally of more than 1,700. Woodward reported back that he had read them all and came back with his top-10 favourites list and Google's response to each, an unusual demonstration of transparency for such a secretive company.
It is certainly not the first time Woodward has crowdsourced his opinions, but this comes right on the heels of the release of Neural Expressive in May.
The number-one complaint was that Google Workspace integrations need to work more reliably, a sensitive spot given how heavily Google markets Gemini as an assistant that moves across Gmail, Docs, Slides, Keep and Tasks.
Woodward confirmed that the Gemini Spark agent had received recent enhancements but conceded that there is still work to be done in terms of improving the overall application's reliability.
The second one was more consistent tool usage, which Google considers an area where it "strongly" agrees with Woodward and will deliver improvements "soon" thanks to future model enhancements. The third improvement would be improved organisation of chat projects and folders, something that Google considers Notebooks to be the beginning of, although not quite up to the mark and needing complete rethinking.
Several other suggestions were considered as currently underway: MCP and Custom Skills, following the recent rollout of the feature in Gemini Spark, allowing third-party integration with Canva, Dropbox, Instacart, OpenTable and Zillow; message editing; improved voice dictation; and mobile scrolling bug fixes.
Deep Research users asked for the ability to export reports to NotebookLM and switch models mid-chat, requests Woodward admitted he hadn't previously realised were pain points.
-
OpenAI, Anthropic and Google join forces on AI safety efforts
-
Meta to deploy custom Arke and Astrid AI chips in 2027: Everything to know
-
Dreamforce 2026: Salesforce unveils Koa AI model and Google Cloud partnership
-
EU Kids Act: New social media rules for under-15s
-
China’s spy chief reveals growing fears over AI’s impact on communist party rule
-
Anthropic's new Claude tool targets financial advisors to streamline compliance and client research
-
Elon Musk’s bold AI safety idea: 'Let rivals test each other’s models'
-
Former Google DeepMind researcher warns autonomous AI could wipe out humanity amid industry-wide safety alarm