Claude Code Is Evicting the Ads From Your TV -- AI Brief October 20

Good day %%first_name%%. People are handing their smart TVs to Claude Code and getting the ads back out, which makes it the most relatable AI story of the week. Also today: Anthropic is reportedly readying a model that guesses what you meant, OpenAI says its newest models can finally read COBOL, Nvidia signs a physical-AI pact with Germany, and CNBC asks who automation hit first.

Claude Code Is Evicting Ads From Your TV TechCrunch

* What happened: TechCrunch reports that owners of Android TV and Google TV sets are pointing AI agents at their televisions to strip sponsored ads, preinstalled apps and tracking. Developer Gathin posted on October 17 that Claude Code found his parents’ TV on the Wi-Fi, removed 25 Xiaomi packages and wrote a new launcher; the prompt became the cleanyourtv project, and he says a few thousand people did the same to their own sets within two days. A second project, dumbtv, has been tested on Fire TV and on Android and Google TV models from Sony, TCL, Philips, Hisense, Xiaomi and Shield.

* Why it matters: Smart TVs run a feature called automatic content recognition, which works out what is on screen from screenshots or audio clips, and TV makers earn money by selling that viewing data or using it to target ads. TechCrunch notes the cleanyourtv method needs no root access and does not block a return to factory defaults, and says the trend could dent the smart TV ad business if it spreads.

* What everyone’s saying: Delight, mostly. Gathin’s original post has about 27,000 likes in TechCrunch’s embed, and the pitch on the dumbtv page is blunt: your TV should run your apps and your HDMI inputs “and nothing else.”

* My read between the lines: A TV maker sells you the hardware, sells your viewing habits, and the only thing standing between you and the sponsored tile is now a chatbot with a Wi-Fi password. It is also the sweetest IT-support story of the year, and worth reading what the agent plans to switch off before you let it loose on Mom’s living room.

Someone finally found a way to make the sponsored tile on a TV go away. Your own to-do list is harder to clean out. Viktor is an AI agent that lives in your team’s Slack and connects to 3,000+ tools, then does the work itself: reports, dashboards, code and campaigns. Not a chatbot. A coworker. New readers get $50 off their first month. Hire Viktor →

Anthropic Readies a Model That Reads Intent The Information (paywalled)

* What happened: The Information reported Monday that Anthropic is preparing a new model that reasons and uses intuition better than its current ones. According to a Sina Finance summary of the report, it is good at working out what users really want from very little information, and at thinking creatively instead of reaching for standard advice. No launch date or name was reported.

* Why it matters: Yesterday we covered Anthropic’s 5% pledge to a public benefit fund; this is the product side of the same pre-IPO stretch. Per The Information, OpenAI, Anthropic and Google have shipped more than a dozen models in the past month, and the pace is not expected to slow, so the model you picked last month is probably not the one on offer now.

* What everyone’s saying: That the race is back on. Reuters reported in September that Anthropic was weighing a new release to counter OpenAI’s GPT-6 Astra, which Ramp data put at about 13% of enterprise AI spending versus about 8% for Claude Fable.

* My read between the lines: Dario Amodei wrote his slowdown essay on September 12. When Anthropic shipped Opus 5.5 ten days later, its own announcement called it “the first model we are releasing since we urged for a slowdown in cutting-edge AI,” and the next one is already queued. Slowdown essay, meet roadmap. And a model sold on guessing what you meant is one to watch for how often it stops to ask.

📖 Further reading: Fable 5 Costs 2x Opus — and Using It Wrong Costs You More Than That — my July piece on the rule I use whenever a new model lands: plan on the best one, execute on the best one you can afford.

The Brief is free and always will be. If you would rather know which new model is worth switching to before you switch, that is the kind of thing the paywalled deep-dives get into, with the full archive behind them. Become a member →

OpenAI Teaches Its Models COBOL OpenAI

* What happened: OpenAI published results on Monday for “low-resource” programming langua