At WWDC this week, Apple finally showed the Siri it has been promising for two years. iOS 27 and the expanded Apple Intelligence stack include an assistant that can act across apps, hold context across a conversation, and start tasks on its own instead of waiting for a wake word and a command. Ten days earlier, on May 30, Google introduced Gemini Spark, pitched explicitly as a 24/7 assistant that watches your day and surfaces things before you ask. And back in April, Google confirmed Gemini is shipping in millions of vehicles, sitting between the driver and the car’s systems.
Three announcements from three companies, and they’re all chasing the same pattern. The thing I wrote about in The Age of the Agentic Loop back in February is no longer an enterprise architecture diagram. It’s in your pocket and your dashboard, and it’s about to be in the hands of a billion people who have never heard the phrase “agentic loop” and never will.
The consumer loop is the same loop
I described the agentic loop as a system that senses, thinks, acts, and remembers, then repeats, without waiting to be asked. I was writing about portfolio monitoring and production alerting and compliance. The consumer version is the same four steps wearing different clothes.
Gemini Spark senses your calendar, your location, your inbox, the traffic on your commute. The revamped Siri thinks across the apps on your phone, deciding that the flight delay in your email matters to the dinner reservation in your calendar. Gemini in the car acts, rerouting you and texting the person you’re meeting. And all three are racing to remember, to hold enough context that tomorrow’s interaction starts from what happened today instead of from zero.
That last step is the one that separates a demo from a product, and it’s the same point I made about the enterprise. Without memory the loop just runs the same play over and over; with it, each pass starts from somewhere better than the last.
Expectations reset for everyone
When the loop goes mainstream, every person who uses one of these assistants for a month is being trained, quietly and for free, to expect proactive behavior from software. And not just from their phone, but from everything.
I have watched expectation resets happen before. At Amazon, working on Alexa, we learned that once people could ask the air to turn off the lights, a wall switch started to feel like a chore. At Drift, conversational AI on a website went from a novelty to table stakes in about eighteen months; buyers stopped tolerating a contact form because they had been taught that a real answer could arrive in seconds. The pattern is consistent: once a capability becomes ambient, people stop noticing it and start noticing only when it’s missing.
So the person evaluating your B2B dashboard in 2027 will have spent a year being nudged by their phone before they asked. They will open your product, wait for it to tell them what changed and what to do about it, and feel a flicker of disappointment when it just sits there with a blank search box. They won’t articulate it as “this lacks an agentic loop,” just register that it feels old.
Reactive products are about to feel dated
For most of software history, the contract was simple: the user makes a request, the system responds. That contract is what is now expiring. There’s nothing wrong with a pull-based, request-response product, except that it’s going to feel like a flip phone sitting next to the thing in everyone’s pocket.
This is uncomfortable because reactive is how almost everything is built. The form, the search bar, the dashboard you have to remember to check, the report someone runs every Monday. All of it works fine on its own terms; the trouble is what it now sits next to.
At Vestmark we have been building toward push instead of pull with Vestmark Pulse, surfacing the account, the drift, the regulatory flag that needs attention before an advisor goes looking. We did that because the enterprise math demanded it. The consumer announcements this week mean the same expectation is now arriving from the other direction too, carried in by every employee’s personal phone.
The differentiator is the context layer
When everyone has access to the same frontier models, and they roughly do, the model stops being the moat. Apple and Google are winning these announcements on context rather than raw reasoning: who has the calendar, the email, the location history, the car, the permission to connect them. Whoever assembles the richest, most trustworthy memory of the user wins the interaction.
That is exactly the bet I argued for in the enterprise. Calling the LLM is the easy part, since we wire RubyLLM into our stack in an afternoon. The real work is the years of accumulated data, integrations, and domain context that make the loop’s “think” step actually smart, plus the identity and permission plumbing (for us, Okta OIDC and a fair amount of careful work on what an agent is allowed to touch) that makes “act” safe to ship. That’s as true for consumers as it is for the enterprise, and the context layer is where the product actually lives.
What to do this week
Pull up the product you own. Find the single most important thing your user currently has to remember to check, the screen they open out of discipline rather than delight. That recurring manual check is your loop candidate, the consumer-facing twin of the “someone checks this every day” test I gave engineering leaders in February.
Now sketch the smallest honest version of sensing it for them, deciding whether it matters, and telling them with enough context that they can act in one tap. You do not need to rebuild the app. You need one proactive moment that earns trust. Ship that, then add the memory that makes the second moment better than the first. Do it before your users spend another year being taught by Apple and Google to expect it, because by then nobody will call it innovative. They’ll just expect it, and notice when it isn’t there.