There's a trend for journalists to approach new technology with a sense of cynicism. It comes from that predisposition to look beyond the marketing to what lies underneath. That left me with serious doubts about Google's fancy new Gemini Spark.
Google I/O introduced the next evolution of Gemini, not just in terms of the underlying language models in Gemini 3.5, but in the consumer-facing entry point. Gemini Spark is described as a 24/7 AI agent, designed to "navigate your digital life".
I followed the Google I/O Keynote in detail and while Gemini was definitely the star of the show (Gemini is basically everywhere) Gemini Spark was the new product that really jumped out.
What is Gemini Spark?
Gemini Spark aims to deliver Google's long ambition to have an "agentic AI". An agentic AI is something that can complete tasks on your behalf, so rather than just asking it to find you a restaurant, it can find that restaurant, book a table and email your friend the details, without you having to open any apps other than the AI app.
"Getting things done" is the mantra of agentic AI, and it's seen as the holy grail of AI, because it takes us beyond rapid information discovery and processing, into action.
This is what Gemini Spark is designed to do: it's designed to work autonomously in the cloud, with access to your Google services. Here's how Google introduced Gemini Spark:
What can Gemini Spark do?
Google outlines that Gemini Spark runs on dedicated cloud servers, so it can be running in the background without you needing your laptop or phone to be connected. That means it can work on things without you being there to prompt it or monitor it.
It works with Google's tools at the moment like Gmail and Google Docs, with third party integrations promised in the future. We already have integration in Gemini for services like Spotify and WhatsApp for example.
The idea is that you can start a Spark task to compile all the information in the last week about an event and summarise it. Rather than sitting waiting for that to happen, that can run in the background while you get on with something else - you'll return to a summary with everything in it.
The second example we've seen is arranging a street party along the same lines, creating a spreadsheet of information from replies that have come in via email - which is all live, because Spark is actively working on that, rather than it being a one-time prompt.
What it's actually doing is creating custom data workflows, but with simple conversational prompts rather than more complicated coding.
That all sounds great doesn't it?
It sounds great for people who want to pull together data from within Google's systems, but it also rings alarm bells. If Spark is doing this compiling for you, where is the accountability? Who monitors the AI to make sure that you're actually getting the results you want, rather than just accepting the results you're given?
Look at it this way: we're all used to driving with satnav and we all agree that it's easier than the previous option, which was driving using maps. In those days you had to plan a route, perhaps write instructions, and then follow them.
The result is that satnav driving is easy - but you never really know where you're going. You don't really know the route, you don't know what's along the route, you're just following instructions and no longer thinking.
Gemini Spark will allow you to stop thinking.
I'm currently planning a family holiday, which is perfect for something like Spark. Indeed, Gemini has already suggested a couple of routes, where to stay, what to avoid and what to do. It was simple - but I have literally no idea what my options are, because I've not actually done any research, I've just been handed a solution.
This is why I have doubts about Gemini Spark as a personal AI. It might be able to keep track of tasks, but as soon as you stop doing that for yourself, you'll only know what Gemini tells you - because you stopped doing the thinking yourself.
