AI Agents, Stolen Jokes, and Sega's Failed Shortcut

The gap between what AI boosters promise and what AI actually ships is getting embarrassing this week.

The "Self-Driving Company" Vision Is Back, and It's Still Vague

Replit's CEO Amjad Masad is making the rounds pitching a future where AI agents — software programs that can take sequences of actions on their own, like an automated assistant that doesn't stop at writing a draft but actually sends the email, books the meeting, and files the report — run the day-to-day operations of a company while humans handle big-picture decisions. It sounds compelling until you ask what "higher-level decisions" actually means for the person whose job currently is the day-to-day. The honest version of this vision isn't liberation from busywork; it's a smaller headcount doing more oversight of systems they don't fully understand. Replit is a coding tool trying to sell itself as a business school. That pitch will land with founders who already wanted to cut staff and just needed a story.

https://techcrunch.com/2026/10/16/replits-ceo-says-the-future-of-work-is-the-self-driving-company/

ChatGPT Can Now See Your Open Browser Tabs

OpenAI's Chrome extension — a small plugin you install in your browser — has been updated to let ChatGPT read the tabs you have open, when you explicitly ask it to. So you could pull up three research articles and a competitor's pricing page, tag Chrome in your ChatGPT conversation, and ask it to synthesize everything across all four tabs at once. A consultant comparing a client's contract terms against three competitor agreements in one pass is a good example of what this actually unlocks. This is a browser extension with access to your open tabs, which may include your inbox, your HR system, or a client's sensitive documents. If you're using this on a work machine where you're logged into client systems, your IT department is probably going to have feelings about it.

https://www.theverge.com/ai-artificial-intelligence/1012510/openai-chatgpt-google-chrome-tabs

Sega Shipped an AI-Upscaled Remaster and Had to Pull It From Shelves

Sega released a remaster of the classic strategy game Total War: Shogun 2 and then had to yank it from sale after players immediately noticed the visuals looked wrong — the telltale sign of AI upscaling done carelessly, where an automated tool increases image resolution without human artists checking whether the results actually look good. AI upscaling isn't inherently bad; it's a tool that works fine with quality control. The problem here was that it appears Sega used it as a shortcut, skipped the review step, and shipped something that customers rejected hard enough to force a delisting. The question worth asking isn't whether someone should have caught this — obviously yes — but why they didn't. A QA team that was under-resourced, a deadline that wasn't moved, a producer who assumed the tool handled it: one of those is the actual story, and Sega hasn't said which.

https://www.theverge.com/games/1012512/sega-total-war-shogun-2-remaster-pulled

AI Personas Are Stealing Comedians' Jokes to Sell OnlyFans Subscriptions

Wired documented a network of AI-generated social media accounts — fake attractive women that don't exist — posting comedians' material verbatim to drive traffic to paid content platforms. The comedians can file complaints, but the accounts multiply faster than platforms can moderate them, and the legal tools for protecting jokes specifically are notoriously weak. This is a specific, nasty flavor of content theft, and it also previews a broader problem that any professional who creates original work will eventually face: AI makes it cheap to launder your ideas through a fake identity at scale. Copyright law was not built for this.

https://www.wired.com/story/ai-generated-hot-girls-are-ripping-off-comedians-to-boost-their-horny-content/

How Cal AI Hit $30 Million by Paying Influencers, Not Engineers

Cal AI, a calorie-tracking app that uses your phone's camera to estimate what you're eating, grew to $30 million in revenue in two years and was acquired by MyFitnessPal largely by building a paid influencer army rather than doing anything technically novel. The AI in the app is real but not unique — computer vision that identifies food has existed for years. What Cal AI figured out is that distribution beats technology in consumer AI right now, because the underlying tools are available to almost anyone. The acquisition is a reminder that in a world where AI capabilities are increasingly commoditized, the actual competitive advantage often has nothing to do with the model.

https://techcrunch.com/2026/10/15/when-it-came-to-cal-ais-viral-success-a-small-army-of-influencers-paid-off/

The Sega story and the ChatGPT tabs story are about the same underlying tension: AI makes things faster, but the judgment about whether the output is actually good still has to come from somewhere, and right now we're figuring out who's responsible when it doesn't.

← More daily briefs