October 6, 2026
Instinct's new group-chat feature lets an AI agent plan a trip or coordinate a carpool with friends who don't even have an account on the platform. On paper that's a convenience win -- no more forwarding screenshots or copy-pasting addresses into five different threads. But look at what Instinct had to build alongside it: a permission layer specifically to stop personal agents from sharing information or taking action without explicit consent. That's not a nice-to-have feature; it's an admission that agents acting on behalf of multiple people, some of whom never opted into the system, is inherently risky. For businesses watching this space, the lesson isn't 'add agents to chat.' It's that every agent capable of taking action on someone else's behalf needs a permission model built in from day one, not bolted on after launch. We've argued before that agents moving into group chats raises real accountability questions, and Instinct's launch is a live test case rather than a hypothetical one.
TikTok's new Shopping Assistant pairs a conversational agent with one-click checkout, meaning a user can discover a product through chat and buy it without ever leaving the app. This matters more than it sounds. Every major consumer platform is racing to insert an agent between the user and the purchase decision -- TikTok is just the latest to do it with real commerce infrastructure behind it, not a demo. The upside for shoppers is obvious: less friction, faster decisions. The catch is that the company controlling the agent also controls what gets recommended, and conversational interfaces make that steering much harder to notice than a ranked list of search results ever was. Any business selling through TikTok, or evaluating marketing channels more broadly, should treat this as a new kind of algorithm to understand -- one that talks back instead of just ranking.</br></br>It's worth putting this next to OpenAI's move to start running visual ads inside ChatGPT's image generation results. Different company, same pattern: the assistant you trust for neutral help is quietly becoming an ad surface. Neither OpenAI nor TikTok is hiding this, which is to their credit, but the direction of travel is clear -- the chat window is becoming the new storefront, and advertisers will follow the attention there just like they followed it to search and social feeds before.
The most unsettling headline today isn't a product launch at all. Independent researchers say they've identified a swarm of AI agents running on Tencent's infrastructure that appears to be targeting Alibaba's map service, Amap. If that holds up, it's one of the first documented cases of an 'agent fleet' operating at scale against a specific commercial target rather than just automating internal tasks. We don't know yet whether this is scraping, competitive intelligence, an attack, or something more benign dressed up as a scandal -- the researchers are still tracking it. But the fact that it's trackable at all tells you agent activity is now large enough to leave footprints across entire platforms. For any business running infrastructure that agents might probe, scrape, or route around, this is a preview of a world where bot traffic isn't just spam bots anymore -- it's autonomous software with goals, operating at a scale that looks like an actual fleet. Security teams should start asking whether their monitoring tools can even tell the difference.
Put these three stories side by side -- agents in friend groups, agents at checkout, agents allegedly probing a rival's infrastructure -- and a pattern emerges. Every one of these launched this year is a company racing to get an agent doing real-world things before anyone has fully worked out who's accountable when it goes wrong. Instinct at least built permissioning in up front, which is more than TikTok's shopping assistant or the Tencent swarm can claim to have shown publicly. That inconsistency is the real story: there's no shared standard yet for what an accountable agent even looks like, so each company is improvising its own answer. Businesses adopting any of these tools should ask vendors directly what happens when an agent acts on bad information, takes an unauthorized action, or gets targeted by another agent -- because right now, the honest answer from most vendors is probably 'we're working on it.'
Which of these worries you more as a business owner: an agent acting on your behalf without full consent, or another company's agent quietly interacting with your platform without your knowledge?
Sources