Hey there,

The last year has brought a shift in AI. What started as a chatbot craze has now expanded to AI agents that can work autonomously in the background. At Google I/O back in May, the Gemini team announced their new personal agent called Spark. It was initially locked to Ultra subscribers, but has rolled out more widely at the end of July. This week we’ll look at what Spark is, how tasks and skills work together, and where you still need to pay attention. 

What is Gemini Spark?

Spark is an autonomous Gemini agent that works in the background on complex, multi-step tasks. Google explains that it works 24/7 and is even able to work while your phone or laptop is closed. Spark doesn’t run locally on your computer and instead runs through Google cloud servers. This allows it to run independently of your machine, so you can start a task and then go offline and take a break.

AI agents are everywhere in 2026 but Spark stands out in a few ways. For one, it is seamlessly connected to your Google account and can access your connected Google Workspace. It can also utilize a number of connected apps.

Spark allows you to explain your goal in plain language, something like organizing a Drive folder, or summarizing your unread emails. Spark will plan out each step, work through them, and then create a complete report of its overall progress as well as a list of every file it interacted with along the way. It's available now to Ultra and Pro subscribers through the Gemini app. 

How tasks, schedules, and skills work together

Gemini Spark has three interconnected pieces: tasks, schedules, and skills. 

Tasks are the prompts you give Spark to complete. These can be anything from “summarize my unread emails for important topics” to “draft a reply using the information pulled from the Google Doc the thread mentions”. 

Schedules build directly off of tasks. They are simply recurring tasks that Spark performs at a set time. These are very similar to “Scheduled Actions” but are much more powerful. These scheduled tasks can be run without your supervision and without you needing to be online. 

Skills are reusable blueprints for how Spark should go about completing a task. Let’s say that you want to create a Quarterly report for work and use Spark throughout the process. Once you’re satisfied with the output and ready to move on, you can have Spark create a detailed guide on how this report was created. All the DOs and DON’Ts are saved as a skill that can be recalled next quarter when you need to create a new version of this report. Simply put, skills allow you to save a process to be reused at a later date. 

What makes skills worth learning

The way skills work is that Spark isn’t reading all the content within your skills and then deciding whether to utilize it or not. It only reads the name and description to start and uses that information to decide whether a skill is relevant to the current task. Only after a skill is deemed relevant will Spark read the full context and start using it. This allows it to only use skills when actually necessary and saves context space. 

This means that your description is acting as a trigger for the skill and a vague description may never be used, no matter how great the instructions underneath it are. Google offers some great advice for creating skills like “Focus on what the skill does, not what it is” when naming it. Or “List specific examples when the skill should be used. Give specific situations where a skill is useful. This tells Gemini when to use the skill.”

If this all sounds familiar, it should! This is the same way that skills work in Claude. In fact, you can share the same skill across both platforms.

My Thoughts

Spark was one of the most exciting announcements from this year’s Google I/O. Having a personal assistant that can help make repetitive daily tasks faster and easier sounds great, but also comes at a price. Like all AI agents, Spark isn’t 100% accurate and can make real mistakes with lasting impacts. Giving it access to your Google account and workspace has massive upsides but also potential dangers. Google warns that Spark is susceptible to hidden prompts that can be embedded within emails, documents, or websites and can override the tasks you’ve assigned your assistant. That means that one of Spark’s greatest selling points, its ability to run while you're away, can turn into its greatest weakness. One plus is that it's running remotely so it can’t access or alter your local files / computer. 

My position on tools like Spark is pretty straightforward. While Spark can work without you, it probably shouldn’t. If the tasks are low stakes, like working within your Drive or pulling together research, it's fine because a bad result just means a few minutes of manual work. However, anything that involves more outward tasks like sending emails or spending money are something I’d still do myself. Despite this, I think Spark is still useful and worth checking out. Just start with the small things and work your way up when you feel comfortable doing so.