Overview
Today’s feed had a strong AI current running through it, from X changing how creators get paid to Google retiring Gemini Gems, OpenAI and Anthropic trading playful blows, and Armenia bringing national AI compute online. There was also room for a Falcon 9 launch, a sharp startup reminder from Paul Graham and Alex Hormozi, and a small piece of telecoms history that still lands with real weight.
The big picture
X moves creator payouts towards original work
X is winding down its existing Revenue Sharing programme, with new sign-ups halted and current participants able to earn until 7 September 2026. From 8 September, the replacement will be Original Content Rewards, with eligibility tied to X Premium, verified followers, and verified impressions.
The main idea is clear enough: stop rewarding repost loops and copied material, and put more weight behind reporting, analysis, original media, and commentary that adds something fresh. Whether creators see that as fair or just another moving target will depend on how the new rules are applied.
Armenia brings sovereign AI infrastructure online
Michael Dell highlighted the launch of Armenia’s sovereign AI infrastructure in Hrazdan, built with Dell PowerEdge servers, NVIDIA Blackwell GPUs, and FirebirdCloudAI. The project gives Armenia local control over high-end AI compute, rather than relying only on foreign cloud providers.
The first phase is reported at $500 million, with approved expansion above 400MW. That would put it among Europe’s largest AI clusters, aimed at demanding work in life sciences, robotics, and space technology.
Google prepares to retire Gemini Gems
TestingCatalog spotted that Google plans to retire Gems in Gemini on 20 October 2026. Users are being told to save their Gem content or recreate it as Skills, which points to Skills becoming the new home for reusable instructions and task setups.
Gems worked much like custom versions of Gemini, with set instructions and files for repeated jobs. Skills look like Google’s attempt to make that idea more modular, so workflows can be reused without starting from scratch each time.
Falcon 9 keeps making spaceflight look routine
SpaceX launched another 24 Starlink satellites from California, using Falcon 9 booster B1093 on its 16th flight. The booster landed on the droneship Of Course I Still Love You, adding another clean entry to SpaceX’s growing launch count for the year.
The striking part is how normal this now feels. Reused boosters, frequent Starlink deployments, and a high success rate have turned what used to be rare national events into regular operating rhythm.
Anthropic’s safety message meets IPO pressure
Polymarket picked up reports that some Anthropic investors want CEO Dario Amodei to cool down his public warnings about AI doom ahead of a possible IPO. It is a neat snapshot of the tension inside the AI industry: safety concerns are central to Anthropic’s identity, but public markets may prefer a cleaner growth story.
Amodei has been open about serious AI risks for years. Now Anthropic is reportedly moving towards a public listing with huge valuation talk around it, which makes every public comment part of the company’s financial story too.
OpenAI and Anthropic’s coding rivalry gets louder
Tibo from OpenAI jumped into the ongoing OpenAI versus Anthropic coding-tool banter by praising GPT-5.6 Sol and saying it can be used across environments, including the CC harness. He also reset usage limits for paid ChatGPT Work and Codex users, which turned a product point into a public flex.
Chubby framed the exchange as more entertaining than the familiar Sam Altman and Elon Musk feud. The fun here is that the rivalry is not just executive drama, it is showing up in developer tools, model access, hiring jokes, and small perks for paying customers.
Invideo adds guided Workflows to Agent Two
Invideo launched Workflows for Agent Two, giving users a guided way to complete creative tasks without spelling out every step. The flow asks questions, takes references, offers choices, and then produces the finished job.
The first set covers filmmaking and marketing use cases, with custom reusable workflows also supported. The demo points towards a more structured future for AI creative tools, where the machine does not just answer prompts, it walks people through decisions.
Simple beats clever in company building
Paul Graham said much of his office hours involve giving founders permission to do something simpler than they had planned. Alex Hormozi came at the same broad idea from a different angle: hold higher standards than customers expect, then charge serious money for the result.
Both posts cut against the founder habit of making things more complicated than they need to be. The useful lesson is plain: reduce the plan, raise the standard, and make sure the business model gives you room to keep quality high.
AI is oddly good at Three.js
The Three.js account asked why AI is so good at producing code for the 3D JavaScript library. The replies pointed to a simple reason: Three.js hides much of the low-level WebGL pain, so models can focus on intent, scene structure, materials, lighting, and interaction.
Developers also shared examples of AI-assisted 3D projects, from FPS-style scenes to richer browser-based environments. It is a reminder that some libraries are friendlier to AI not by accident, but because they are well designed for humans too.
A minute of silence for Alexander Graham Bell
World of Engineering shared a striking historical note: when Alexander Graham Bell was buried in 1922, the US and Canadian telephone systems were silenced for one minute in tribute. Around 13 million phones and more than 60,000 operators were part of the pause.
It is hard to imagine a modern network stopping in the same way for an inventor. That short silence said plenty about Bell’s place in the history of communication, and about how visible the telephone’s transformation of daily life already was.
