Meta's AI Agent, Google Pays Publishers, and AI's Scaling Limits

The pattern today is the same across every story below: the companies building AI tools and the companies building the pipes those tools run on are all making the same ask — more access, more trust, more lock-in. Whether the value they're offering back is worth it depends on which story you're looking at.

Meta's New AI Agent Will Run Your Life (If You Let It)

Meta launched Muse, an AI agent that can write your emails, make purchases on your behalf, and handle other tasks you'd normally do yourself. An AI agent, if you haven't encountered the term yet, is different from a chatbot — instead of just answering questions, it takes actions in the world on your behalf, like clicking buttons and completing transactions. Muse works, apparently, and sometimes works well. The catch is that to do any of the useful stuff, you hand Meta your credit card and, implicitly, a much richer picture of your behavior than even your Facebook feed already provides. Meta already knew what you clicked on. Now it wants to know what you buy, what you delegate, and how you make decisions. That's not a fair trade — it's a significant one, and most people will agree to it without thinking twice.

https://www.theverge.com/ai-artificial-intelligence/1002671/meta-muse-ai

Google Is Testing Payments to Publishers It's Been Quietly Starving

Google has started paying a small group of roughly 100 publishers whose content feeds its AI-powered search answers — the summaries that appear at the top of a search result page, often replacing the need to click through to any website at all. AI summaries have been eating into publisher traffic for two years, and this pilot is Google's first acknowledgment that the arrangement has a cost someone else is bearing. The number to watch is what percentage of the value Google is actually sharing. If it's anything close to what display advertising used to pay out, this is a real revenue model. If it's closer to what YouTube pays per thousand views on a niche channel, it's a goodwill gesture designed to buy time while Google keeps the rest.

https://www.theverge.com/tech/1002665/google-paying-publishers-ai-search-features

Instagram Wants to Tell You How to Post Better

Instagram is adding an AI creative assistant to its Edits app — a standalone editing tool Meta released earlier this year — that will analyze your account's past performance and suggest how to improve your posts. Based on what Meta has described, it will surface patterns like which formats drove the most reach and how caption length affected saves. That's useful information, and creators who currently pay for third-party analytics tools will likely find it genuinely helpful. The more honest framing, though, is that Instagram is building a coach whose job is to keep you posting more, more consistently, and in formats the algorithm rewards. That's not necessarily bad for you — but it's not neutral either, and treating it like a disinterested advisor would be a mistake.

https://www.theverge.com/tech/1002402/instagram-ai-edits-assistant-creators-analytics-youtube

The Unglamorous Infrastructure Bet That Could Matter More Than Any of the Above

A startup called Restate just raised $20 million to build what's called "durable infrastructure" for AI agents — meaning software that ensures an AI agent's tasks don't fail silently if something crashes or a connection drops midway through. Think of it this way: if an AI agent is booking a flight for you and the system hiccups at the payment step, durable infrastructure is what decides whether the booking completes correctly or leaves you with a charge and no ticket. It's boring, and it will probably determine whether enterprise AI agents are actually deployable at scale or stay permanently in the "promising pilot" category. The founding team comes from Apache Flink, which means they've already built systems designed to process millions of data events per second without dropping records — the failure modes they're solving for at Restate are a direct extension of that problem, applied to agent workflows instead of data pipelines.

https://techcrunch.com/2026/09/30/restate-lands-20m-as-the-need-for-durable-infrastructure-increases-with-ai-agents/

Can AI Keep Getting Smarter? A Hardware CEO Says Maybe Not Forever.

At TechCrunch Disrupt 2026, Cerebras Systems CEO Andrew Feldman is scheduled to speak about what happens when the hardware powering AI models hits its physical and economic limits — a question that gets far less attention than it deserves given how many business roadmaps are quietly built around the assumption that AI capabilities will keep compounding. Cerebras builds specialized AI chips — processors designed specifically to run AI workloads rather than general computing tasks — and Feldman has been publicly skeptical about whether the current approach to scaling AI (bigger models, more chips, more power) can continue indefinitely. If he's right, the next leap in AI capability won't come from just throwing more resources at the problem, and that changes the timeline most businesses are quietly planning around.

https://techcrunch.com/2026/09/30/cerebras-systems-andrew-feldman-on-whether-ai-can-keep-scaling-at-techcrunch-disrupt-2026/

The hardest thing to know right now is whether you're watching the early innings of something that keeps compounding, or the part of the movie where everyone's still confident right before things get complicated.

← More daily briefs