Last week, Google rolled out Gemini Spark: A 24/7 personal AI agent that lives across your Google account, reads your stuff, takes action on your behalf, and runs in the background even when you're not asking it to.
It's available to Google AI Ultra subscribers in the US starting now.
In today’s edition, we’re talking about why this is the first AI personal assistant that has a real shot at actually working for most of us.
Let’s get into it 👇
Your best prompts are the ones you'd never bother typing.
The detailed ones. The ones with examples and edge cases. Wispr Flow lets you speak them instead — clean, structured, ready to paste into any AI tool. Free on Mac, Windows, and iPhone.
The PA dream
Every AI lab has been chasing the same vision: a personal AI assistant that handles your busywork, manages your calendar, drafts your emails, plans your trips, and ultimately saves you hours every week.
AI models (ChatGPT, Claude, etc) have been good enough for over a year.
The hard part is getting those models access.
An effective AI assistant needs to know what's in your inbox. Which meetings you've accepted. Which docs you've been editing. Who you email the most. What's on your calendar next Tuesday at 2pm. Without you having to wire any of it up.
That's the moat: context.
And no one has that context like Google.
Why Spark is uniquely positioned
Spark plugs into:
Gmail (read, search, draft, reply, label, summarize)
Calendar (check, schedule, reschedule, RSVP, find slots that work for multiple people)
Drive, Docs, Sheets, Slides (search, edit, create new from scratch)
A remote browser (navigate to sites, click around, add to cart)
Your location, sign-in history, and personal context across Google services
You authorized all of it the day you made a Google account. Spark inherits that access the moment you switch it on.
It's just there.
Other assistants can get to similar places:
ChatGPT has connectors.
Claude has MCP servers (a standard way to bolt an AI onto your apps).
Plenty of agents can hook into Gmail and Calendar with the right plumbing.
But plumbing is the operative word.
There's friction at every step: authorize this, allow that, refresh the token when it breaks.
Google skips all of that. They built the productivity stack and they own the agent.
Not to mention, most knowledge workers already live in Google.
Gmail. Calendar. Drive. Docs. The footprint is enormous.
Spark doesn't ask anyone to switch tools or learn a new workflow. It turns the workflow you already have into one that acts on its own.
The AI assistant race isn't being won by the smartest model. It's being won by the assistant with the most context already plugged in.
How Spark thinks about work
Three pieces worth knowing before you start using it.
Task. What you want done. "Plan my business trip to London." "Give me a morning brief." "Track the AI news from this week."
Schedule. When the agent runs. Could be a specific time (every weekday at 8am). Could be event-driven (when my flight is delayed).
Skill. A reusable recipe for how to do a kind of work (think of it as a saved playbook the agent follows). Things like "Travel Booking," "Email Drafting in My Voice," "Weekly Newsletter Triage." Build it once, Spark uses it whenever it applies.
Task = what. Schedule = when. Skill = how.
If "Skill" sounds familiar, it should. We covered this exact concept a few editions back when we talked about turning your automations into reusable tools.
Google just made it a first-class citizen in their OS-level assistant.
If you needed further illustration, here’s an example workflow (for booking travel) that Google provides:
Task: Plan and manage my business trip to London.
Schedule: When my flight is delayed, notify me and propose an itinerary update.
Skill: Use the "Travel Booking" skill and the "Gmail Writing" skill together to rebook my room and send a confirmation.
What to do with it on day one
If you've got Ultra and you're starting from scratch, three tasks worth trying first:
1. Morning brief. Schedule it for 8am every weekday. "Give me a one-screen briefing: today's weather, top 3 calendar priorities, any urgent emails I haven't responded to, and one thing to make today better."
2. Inbox triage. Schedule it weekly. "Archive newsletters older than 7 days. Unsubscribe from any list I haven't opened in the last 30 days. Surface anything that looks like it actually needs a reply."
3. Meeting scheduler. One-off task. "Find three 30-minute slots next week that work for me and [name]. Draft an email proposing them."
Each of these saves real time. Each is also a good stress test of what Spark can handle today (it's still in beta, still maturing).
Two honest tradeoffs
It needs Ultra. This isn't free. The PA tier is premium for a reason: it burns compute around the clock.
It has access to a lot. Your email, calendar, docs, browsing. Worth thinking carefully about what skills and schedules you set up, especially anything that touches money, sensitive contacts, or anything you'd want to approve before it goes out.
The bigger picture for AI assistants:
Every lab is going to land at roughly the same model quality over the next year.
The differentiator will be how much of your life the assistant already touches the moment you log in.
Spark is Google's first real attempt to turn 20 years of accumulated context and integrations into something that works for you instead of just sitting under you.
The other labs will get there too. But Google has a head start on the data, the integrations, and the workflows people already use every day.
Hard to bet against that!
Spark requires Ultra, but you don't need Spark to figure out what you'd use it for.
This week's prompt has your AI of choice (Claude, ChatGPT, Gemini, anything) help you identify your three highest-value PA automations 👇
Help me figure out the three best tasks I could hand off to a personal AI agent.
I'll describe how I spend a typical week. Ask me clarifying questions one at a time until you have a clear picture of what's eating my time. Then suggest three specific tasks an AI agent should handle for me.
For each suggestion, include:
- What the agent does
- When it runs (daily, weekly, on a trigger)
- What it needs access to (email, calendar, etc.)
- What "done" looks like
Aim for tasks that are repetitive, well-defined, and would save me at least 15 minutes a week each.
Ready when you are. Ask me your first question.
Two notes:
Save the output. Whatever the agent suggests, you can use those exact specs in Spark (or Claude with MCP, or ChatGPT with connectors, or any agent that can hook into your tools).
Start with one. Don't try to set up all three the same day. Pick the highest-value one, get it working for a week, then add the next. Reps compound.
Your roundup of the latest model releases and updates from the biggest AI labs.
Claude Opus 4.8 lands, tuned for honesty (May 28). Alongside the raise, Anthropic shipped a new flagship model. The headline improvement is reliability: it's four times less likely than the last version to hand you broken code without flagging the problem itself. For anyone using AI to draft or build, a model that catches its own mistakes is worth more than one that's marginally smarter. (SiliconANGLE)
Microsoft makes Agent 365 generally available (May 1). This is the enterprise mirror image of Spark: a control panel for companies to deploy, watch, and lock down fleets of AI agents the way IT already manages employee laptops. It's $15 per user a month. The real tell: "your whole company runs agents" is now a line item IT departments are budgeting for. (Microsoft)
OpenAI gives ChatGPT a memory of your inbox. ChatGPT can now pull context from your past chats, your files, and a connected Gmail account to personalize what it does for you — and it shows you exactly which sources it used, so you can correct or delete anything stale. Same race as Spark, same prize: the assistant that already knows your context wins. (OpenAI)
Advertise with Build with AI
Get in front of an audience of professionals using AI day-to-day: founders, engineers, operators, and product builders.
Interested in advertising? Respond to this email for rates and details.
Until next time,
William Ryan
Editor-in-Chief @ Build with AI
PS: Follow me on X for daily updates and AI workflows.