PROJUKTI BRIEF
1 OpenAI moves into the space Anthropic left behind as Mythos stays dark.
It has been more than 10 days since Claude Mythos and Fable were pulled, with The Economist reporting that Mythos broke into nearly all of the NSA's classified systems in hours during a government exercise.
While Anthropic works through negotiations to release its models, OpenAI has moved quickly to fill the vacuum: expanding its cybersecurity initiative by rolling out the full version of GPT-5.5-Cyber to trusted partners and launching Patch the Planet, a program designed to help organizations find and close vulnerabilities before someone else does.
2 ElevenLabs launches an engine that auto-translates your ads into 50+ languages.
ElevenLabs Ads Engine connects to your Meta, Google, and LinkedIn accounts and handles the full translation of existing ad creatives across more than 50 languages, adapting text, images, and video dubbing automatically. It then publishes the updated versions and pulls live performance data to flag when an ad starts losing momentum.
For any brand running multilingual campaigns, the manual version of this workflow currently takes days. See how it works.
3 Nvidia's zero-water AI factory design picks up 12M views and renewed momentum.
AI's water and energy demands have become a growing point of friction for data center projects. Nvidia's answer is a factory design using a closed-loop cooling system that achieves near-zero water consumption in favorable climates. First unveiled in early June, the design resurfaced this week after a viral post racked up 12M views and put it back in front of the people building the next generation of infrastructure.
PROJUKTI FORWARD
Two builders agree. Boris Cherny, the creator of Claude Code, and Peter Steinberger, founder of OpenClaw, have independently landed on the same prediction: the next major productivity shift in AI will come not from better models, but from loops. That kind of convergence between builders who think deeply about this every day is worth paying attention to.
What looping actually means. Instead of treating your AI like a coworker you chat with, looping asks you to act like a manager. You set a goal with clear instructions and benchmarks for success, and the agent executes a plan, checks its own work, iterates, and verifies the output against your definition of done. Once the loop is configured, it can run the same workflow repeatedly without you in the room.
Why the verification step matters. The part that most people skip is what makes looping either work or fail entirely. Without telling your agent what success looks like, a looping workflow can run indefinitely, produce outputs that look complete, and burn through your usage limits while delivering nothing you can actually use. The verification step is not optional. It is the whole point.
Not for everything yet. Looping still requires enough technical confidence to set it up correctly, and the compute costs can add up fast on complex tasks, so it is not the right fit for every workflow right now. But the concept is spreading quickly. This viral essay covers everything you need to know about loops and has already reached 7M views, which suggests the audience for this idea is much larger than the people currently using it.
PROJUKTI SNAPSHOT
Ditch the keyboard. An OpenAI engineer argues that editing prompts by hand is the slowest way to work with AI, and recommends switching to your AI's built-in dictation feature instead. His technique has collected 2,000 bookmarks and takes about two minutes to adopt.
Getting into a frontier lab. Vlad Feinberg published a blog post on what it actually took to earn a seat at Google DeepMind, and Business Insider featured it shortly after. If you have ever been curious what separates the people who land those roles from everyone else who applies, the answer is in there.
Dan Koe on surviving AI. Writer Dan Koe followed up his widely-read "How To Fix Your Life in One Day" with a new essay: How to Survive AI Mass Replacement. It has pulled 1.5M views and the framing is deliberately provocative, which is exactly why it is worth reading before you have an opinion on it.
AI fact-checker for politicians. A new Chrome extension uses AI to fact-check politicians in real time during live debates, interviews, and speeches. Try the plugin here and form your own view on whether you want it running.
reCAPTCHA is changing. AI has gotten good enough at identifying stoplights and crosswalks that Google is reportedly planning to replace its "prove you're human" test with something harder and more time-consuming. The post has 1M views, which is a reasonable measure of how much everyone is already dreading what comes next.
Meet Elias Thorne. Ask almost any AI to generate a story and there is roughly a 25% chance the protagonist will be named Elias. The pattern is real and unexplained, and it points to something genuinely interesting about how language models develop biases that no one fully understands yet.
