The consumer agent just started to feel real.

I told Instinct: “You know that thing we did? Move it back 90 minutes.” It remembered the booking, the pub’s website, changed the reservation and sent me the confirmation.

Most businesses are not ready. https://t.co/B7glREWkI9

Instinct Made the Consumer Agent Feel Real

It isn’t quite the consumer inflection point. But it is the first personal agent that feels like a product instead of a project—and it makes the next channel shift impossible to ignore.

“You know that thing we did? Move it back 90 minutes.”

I said that to Instinct this morning.

And it knew exactly what I meant.

In case you missed it, Instinct is a personal AI agent with no app to learn. You text or voicenote it. It remembers the context of your life, then uses websites and apps to get things done. It has taken the VC world (and my X feed) by storm because it keeps producing the moments other AI agents promised: “Oh wow, it did it. It just worked.”

How did it know what “that thing” was?

A few days earlier, I had asked it to book Sunday lunch for me and the family. I’d been away for a few weeks, so getting everyone together was quite important to me. Instinct found the local pub I wanted, dealt with its slightly idiosyncratic web form, and made the booking.

Then we had to move it back a day. So I sent a voice note saying, “Can you move it back a day?” It changed the booking and the confirmation arrived. Then this morning, I had to move it again. And it just worked.

It has also booked cycling lessons for my daughter, made better sense of my upcoming week than a professional EA or my ChatGPT with calendar access did, and, when it could still read WhatsApp messages, helped me make sense of what other people had said (Thanks to high masking autism; I struggle with that sometimes :)).

Other early users are seeing the same thing. In five days, Sheel Mohnot used Instinct to find an in-network podiatrist and fill in the forms, negotiate a Comcast bill from $100 to $60, book a DMV appointment, and link two separate United reservations. None of those tasks is life-changing. Together, they make hours of life admin disappear.

I’m starting to just give it the stuff I was dreading.

Instinct feels qualitatively different to anything that came before it. I’ve been writing about consumer agents for a long time. I use OpenClaw. I’ve written about the death of apps, and the UI, but this is the first time I felt it.

If you haven’t used it, you probably won’t know.

A few people have said to me that “ChatGPT could do this already.” And yeah, sure, kind of. A bicycle and a rocket can both move a human from A to B. That description is technically true and completely misses the product. The difference is not the capability list. It is that I can speak in fragments, trust the memory, and watch the confirmation arrive.

So many little mind-blowing moments add up to one big aha moment.

Agents are the new channel. And everything about how consumers and businesses interact is about to change.

The product has finally been packaged

Until now, personal agents were expert mode.

You bought a separate Mac Mini, installed OpenClaw or Hermes and spent multiple weekends getting the setup right. You gave it access to things, debugged it, built the memory around it, and tried to stop context dropping out at exactly the wrong moment. Occasionally useful, often frustrating, and absolutely for hobbyists only.

The model intelligence isn’t the issue in a lot of “agent” experiences. The hard part is building context: helping the agent remember who is who, what happened before, and how you like things done. Every interaction was supposed to make the agent better, but mostly you spent your life explaining yourself again in every new chat window.

Instinct has packaged an ability to remember your idiosyncratic context into a consumer-grade experience. Its ability to remember the people and loose threads in your life is frankly astonishing. Grok Bot is doing something similar for a slightly more technical audience: people who never went near OpenClaw are spinning up five or six bots, giving them jobs and enjoying the product. The main complaint is that they keep hitting utilization limits.

That tells you a lot.

Neither is quite the consumer inflection point. Instinct still breaks. Grok Bot is still fiddly. But they are pointing at what it could be, and we are suddenly much closer than most businesses realize.

Agents as UI will be bigger than mobile

I think this is bigger than mobile.

And in a lot of ways, I built my career during the mobile wave. My first big break was driving the launch of the second mobile banking app in the UK market for TSYS and Capital One back in 2010. Back then, the ability to execute a mobile experience and get it to the app store was everything.

In the 16 years since (god, 16 years, I’m old), everything has become mobile-optimized and mobile-first. Every previous consumer channel changed how a business reached you. The same is about to happen for agents. Every channel reset what it meant to be available—and who got to stand between the business and the customer.

But the human still did the work.

The human walked into the shop, opened the website, and tapped through the app. Mobile made the interface more convenient, but you still had to operate it. With an agent, you describe the outcome and the software goes to the business.

That is the shift I had understood but not felt. Instinct didn’t make the pub’s booking form better. It made the booking form irrelevant to me. The customer is still human. The lunch is mine (and it was still tasty). The family context is mine. The money and the decision are mine. But the user of the pub’s website was software acting for me.

Everything you thought about channels changes when the customer sends software instead of arriving themselves.

I want my agent to use your thing

There was a great line on X:

“I don’t want your agent. I want my agent to use your thing.”

That is the business consequence you have to internalize.

Every company is building an agent inside its own product. Your bank wants you to use its agent. Your airline wants you to use its agent. Your CRM wants you to use its agent.

But I don’t want to explain myself separately to 50 company agents. I want the agent that already knows me to use those 50 companies.

That’s why I think what Marc Benioff is doing with Salesforce and Anthropic is brave. Claudeforce puts Salesforce’s live data, workflows, and business rules inside Claude. A seller can reason over pipeline, update records, and take governed action without opening Salesforce.

Salesforce is recognizing that it can remain the system of record without owning the intelligence or the customer’s primary interface. It needs the customer’s agent to be able to use its thing securely.

Most consumer businesses have not made that inversion.

Today my agent can’t use your thing because you keep breaking your thing because you don’t want agents to use it.

The website sees a bot crawling over it and goes: “No thank you.”

Historically, that made sense. Bots scraped your content, hammered your inventory, stole prices and tried to take the customer relationship away. Certain airlines are especially aggressive about blocking them.

But the same bot-shaped traffic may now be the customer’s representative arriving with intent, permission and money. The website thinks it stopped a crawler. The customer thinks the website is broken.

Give the agent a way in

Companies have spent the last two decades optimizing their entire funnel for online and mobile. They’re about to have to do that for agents, and it starts by giving the agent a reliable way to act.

WebMCP is one early, still-draft way to do that. Non-technically, it lets a website tell an agent what it can do. Instead of the agent staring at pixels and guessing which button changes a booking, the business can expose a reliable “change booking” action.

It gives the agent a door.

Being callable is only the first layer. The full agent-acceptance stack is identity, authority, payment, liability and recourse: can the business know which human is behind the agent, what it may do, how it pays and who unwinds the mistake?

Salesforce is making the system of record available to somebody else’s agent. WebMCP makes websites callable.

Speaking of payments: Link Stripe’s consumer wallet now gives Instinct a safer way to spend. It is US-only for now.

Noah Shinn@noahrshinn·Aug 29Instinct + Stripe

Instinct users who make purchases spend over $1300/month on average through Instinct. Using Instinct to make high value purchases is a sign of trust that we deeply value.

We’re partnering with Stripe to deliver a more seamless payment experience. Instinct hasShow more881161K477K

Instinct finds the thing. Link sends the user a request showing the merchant and amount. The user approves it in the Link app. Link gives Instinct a one-time-use card authorized for that amount, funded by a card or bank account already saved in the wallet. Instinct can pay without ever seeing the underlying credentials.

That sounds like a small product feature. It isn’t. One of the scary things about Instinct was handing a young company your passwords and card details. Link separates permission to spend from possession of the card.

Now, anywhere cards work, boom, your agent can buy things.

Who will be the wallet that holds all of your cards for your agent? Stripe got itself into that position almost immediately. I imagine others will get there too, but whoever starts to win the default wallet for agents has a unique advantage. That wallet can sit between the agent’s intent and millions of businesses that already accept cards.

That is what being ready for a new channel looks like.

I don’t think you’re ready

(Fo’ this jelly)

We had a lot of hype about agentic commerce, which so far turned out to be worse than regular commerce. But clearly, what we were missing was an agent experience that felt like magic. We now have that.

I don’t know when the consumer inflection point comes.

Instinct isn’t quite it. Grok Bot isn’t quite it. OpenAI, oddly, has pivoted hard towards coding and business users as their early attempts definitely weren’t it. (Fidji Simo reportedly told staff not to let OpenAI get distracted by “side quests.” OpenAI itself now says enterprise is more than 40% of revenue and on track to match consumer this year.)

In a lot of ways, this consumer agent really should be Meta, Google or Apple. If Meta acquired Instinct and baked it into WhatsApp, that would be incredible: the consumer system of record, the interface and the agent in one place.

But crystal ball gazing aside, I can now see the product. More importantly, I can feel why people will want it.

Once consumers have an agent that remembers the context, holds their identity, and can act with permission, they will not want to go back to finding the app, recovering the password, searching the inbox, and filling in the form.

There are still all these little breaks. The security model is unfinished. The standards are early. Websites fight the agents. Sometimes the whole thing fails and hands the job back to you.

But the direction is suddenly obvious.

Before you commission another chatbot, try one thing. Give a trusted consumer agent a real job on your business: find something, change it, cancel it, and pay for it. Watch where it breaks. Every failure is a piece of your next channel strategy.

What happens when every consumer has an agent and your business gives them a terrible experience because that is the channel they chose to use?

Can my agent use your thing?

If it can’t, whose thing will it use instead?

ST.