TLDR 2026-01-14

Apple recently announced that it had decided to rely on Gemini to power its AI features. The company will be able to fine-tune the model so that it responds in Apple's preferred manner. Apple's AI won't respond with any branding related to Google or Gemini. Some of the features will launch this spring. More advanced features are expected to be announced at Apple's annual developer conference in June.

Meta and EssilorLuxottica are discussing potentially doubling production capacity to 20 million units or more by the end of the year to capture growing demand and head off rivals. They may potentially further establish the capability of producing more than 30 million units if demand justifies such a move. No final decisions have been made yet. The talks signal confidence that smart glasses can move beyond early adopters and reach mass-market scale.

China's Institute of Radio Spectrum Utilization and Technological Innovation has filed a request with the International Telecommunications Union to operate a satellite constellation with a total of 193,428 satellites. The request far exceeds the ambitions of other current constellation projects. The Chinese constellations plan to use a wide range of radio frequencies and will be deployed across numerous orbits from 300 to 600 kilometers, as well as higher objects at around 20,000 kilometers. The US Federal Communications Commission views China's satellite ambitions as a national security threat.

Reports from a supply chain leaker indicate that OpenAI's first AI gadget is aimed at taking on Apple's AirPods. The device, codenamed 'Sweetpea', is an audio product that resembles an earpiece. It is projected to be released around September. OpenAI has other hardware in the works, including a home-style device and a pen.

Ramp adopted the Builder mindset and an ambitious plan: to become the most productive company in the world. So far, they shipped 270 features in H1 2025 (more than all of 2024 combined), and 90% of 1,200 employees use Notion AI monthly. Read more on the Notion blog →

Google deprecated JPEG-XL support in Chrome/Chromium back in 2022. The decision was widely criticized, and Google said that it may reconsider. The code to restore support was merged this week, so JPEG-XL image support is back in the very latest Google Chrome/Chromium codebase.

Smart specs guide agents clearly, stay within practical context sizes, and evolve with projects. Aim to create clear specs that cover just enough nuance to guide the AI without overwhelming it. Break large tasks into smaller ones. Plan first in read-only mode before executing, and iterate continuously. This post presents a framework for spec-writing that keeps AI agents focused and productive.

The tech world is moving towards voice-powered computing. Several companies are building purpose-built hardware for AI that will be predominantly controlled via voice. This will help set up the next wave of technology: robots. Robots will be wholly voice-controlled.

AI should be part of every process, but the final filter must be human judgment informed by taste. The hardest decisions aren't reducible to rules. AI can tell you whether something works, but only taste can tell you whether it belongs. Personal taste is the moat when execution is cheap and correctness is abundant.

The partnership gives Apple time to manage expectations across its platforms as it develops its own in-house AI server chips, expected to enter mass production in the second half of 2026.

Confer is an open source AI assistant that provides strong assurances that user data is readable only to account holders.

Today's masses still have power due to their labor contribution and their ability to physically rebel against the elite, but AI has the potential to change that power balance.

VoidLink features more than 30 modules that can be used to customize attacks.

Headroom is a context optimization layer for LLM applications that cuts costs by 50% to 90% without losing accuracy.

Early-stage founders who think they have engineering management problems should go back to building product and talking to users.