Under Your Direction | Eliran Keren

Google launched Gemini Spark this week.

$100 a month. Runs in the cloud around the clock, even when your phone is off. Connects to Gmail, Calendar, Drive, Docs, Sheets, Maps. Books the flight, schedules the meeting, updates the doc, replies to the email.

Sundar Pichai introduced it with a specific phrase: “your personal AI agent that helps you navigate your digital life, taking action on your behalf, and under your direction.”

Three words are doing all the work: under your direction.

What “under your direction” actually costs

I want to be precise about what Google built. This is real infrastructure. A 24/7 cloud-native agent that takes multi-step action across your applications — not a chatbot wrapper, not a demo. The execution layer for your digital life, priced and packaged as a consumer product.

That matters.

But Pichai’s phrase contains a constraint nobody in the I/O coverage is naming: under your direction.

Direction is not free. It is not fast. It is not available 24/7.

Direction is the cognitive work of deciding: what matters right now, given everything that’s open, everything that’s changed since yesterday, everything you committed to last week that still hasn’t closed? That question is answered before you give Spark any instruction at all.

A 24/7 execution layer that requires your direction on demand doesn’t reduce that cognitive work. It exposes it.

Execution was never the expensive part

There’s something worth examining in how Gemini Spark is positioned: as relief from execution burden. You don’t have to book the flight yourself. You don’t have to draft the follow-up email. You don’t have to manually push the meeting.

None of those were the expensive part.

The expensive part is what happens before you give Spark any of those instructions. You had to notice the flight needed booking. Notice the follow-up was overdue. Decide the meeting matters and that it should move, and when it should land. That cognitive work — the scanning, the prioritizing, the deciding what the open state of your life actually requires right now — is untouched by a faster execution layer below it.

You’ve hired someone excellent at doing exactly what you tell them. You’re still the person who has to figure out what to tell them.

For $100 a month, Spark makes execution essentially free. The cost it doesn’t change is the one you spend before you give any instruction at all.

What an always-on agent creates

Here’s the part nobody is saying: always-on execution might make the direction problem worse.

When you have a 24/7 capable agent waiting for instructions, something shifts about how you experience open loops. The friction of the unfinished task used to include the friction of doing it yourself. Now it’s different. The unfinished task sits next to an agent that could handle it in thirty seconds — if you’d just give the instruction.

The loop stays open not because execution is hard. It stays open because you haven’t decided what to do. Spark is ready. You’re the bottleneck.

That’s not a failure of the product. It’s an accurate description of where the missing layer is.

The layer Spark doesn’t have

What sits above execution is not better execution. It’s a system that knows what’s actually open.

Not a to-do list. Not a calendar. A system that has processed the context — the email thread from Tuesday, the decision made Monday that nobody has acted on, the commitment made in the meeting that hasn’t been tracked anywhere — and can tell the execution layer what to do without asking you first.

Not “under your direction.” More like: here’s what needs to happen today, based on what I know about what’s open and what matters. Already handled: the follow-up from Thursday, the reschedule you were putting off, the three things that were waiting on a reply that came in overnight. Needs you: the decision about the proposal, because I don’t have enough context to move it.

One layer runs tasks. The other knows which tasks to run.

What the phrase actually tells you

“Under your direction” isn’t a flaw in the product copy. It’s an honest engineering statement.

Building a system that takes action across your apps at the speed and scale Gemini Spark promises requires solving enormous technical problems. Google solved them. What they described accurately is the boundary of what they built: reliable, capable, multi-step execution — contingent on direction.

That’s the right product for the layer they’re building.

It also draws the line exactly where the next category begins.

Gemini Spark is the execution layer. What comes next is the system that doesn’t wait for direction — because it already knows what matters, what’s open, and what changed while you were asleep.

Not because you asked. Because it knows.

That’s not a feature. That’s a different kind of infrastructure.

Eliran Keren — Founder of Deeplica, building the coordination layer for knowledge work.

Sources: Google Gemini App Blog — The Next Evolution · TechCrunch — Google Introduces Gemini Spark · PCWorld — Google’s new Spark AI agent