In partnership with
AIHOY and happy weekend, AInauts!
Maybe you missed a few exciting AI news stories, tools, or practical hacks last week, or perhaps you just joined us. No need to worry!
We rounded up our highlights from the past week:
We’ll finish with the most important Quick News so you can scan every relevant development in one place. Ready? Let’s go!
Write docs 4x faster. Without hating every second.
Nobody became a developer to write documentation. But the docs still need to get written — PRDs, README updates, architecture decisions, onboarding guides.
Wispr Flow lets you talk through it instead. Speak naturally about what the code does, how it works, and why you built it that way. Flow formats everything into clean, professional text you can paste into Notion, Confluence, or GitHub.
Used by engineering teams at OpenAI, Vercel, and Clay. 89% of messages sent with zero edits. Works system-wide on Mac, Windows, and iPhone.
An AI employee can now rebuild a website from an existing URL: carry over the content, develop the design, write the code, and even connect the domain.
Our test took 61 minutes from approval to a verified live site.
The catch is the brief. A vague “make it look good” still produces the purple card-heavy slop everyone recognizes. A better process starts with the original site or a short interview, a clear visual direction, and an agent with a measurable goal.
200+ Proven Ways to Make Money With AI in 2026
The next wave of millionaires will be people who figured out how to make AI work for them.
The window to get ahead is still open. But not for long.
Here are 200+ proven ways to make money with AI in 2026.
Sign up for Superhuman AI, the free daily newsletter read by 1M+ professionals, and get instant access to all 200+ ways to profit from AI this year.
ChatGPT collects facts, preferences, and old goals about you. Some are accurate. Others are assumptions that eventually get treated like fixed rules. Then the answers start drifting, and you blame the model.
Our memory-check prompt helps you separate facts, assumptions, and uncertainty. It takes a few minutes and improves every conversation that follows. Small maintenance, big leverage.
The new Kimi K3 model offers frontier-level intelligence, processes enormous context windows, and is expected to be released as open weights. You probably will not move your entire ChatGPT workflow because of it. But it changes the market: top-tier performance keeps getting cheaper, and it no longer automatically comes from California.
Did you know Claude Cowork can watch you work and turn the process into a reusable skill that handles the task for you next time?
You complete the task once while explaining the clicks, exceptions, and intent. Your AI employee then understands the workflow. That is much easier than writing a long process document, and genuinely useful in everyday work.
AI video gets expensive quickly when every clip is generated from scratch and takes three attempts. Magnific combines stock media, current image and video models, upscaling, and vectorization in one platform. Our agent completed four tasks through the connector without forcing us through multiple interfaces.
The practical approach: use stock for anything that already exists. Generate only your product, character, or a scene you cannot find elsewhere. That saves credits and time.
New transparency obligations under the EU AI Act take effect on August 2. For ordinary text, the situation remains relatively straightforward when a human reviews the content and takes responsibility. Labeling matters more for realistic AI-generated images, video, and audio. This is not legal advice, just a reminder to check what you need to label visibly.
AI News Quickie: The Week’s HAI-lights
AI never sleeps! OpenAI connects ChatGPT to health data, GitHub lets agents work from Linear and Issues, Claude learns from screen recordings, and Google shows how people really use AI day to day. Security, money, regulation, and the growing compute bill also take center stage.
Language note: links to AInauten News open German-language reports; all other sources remain in their original language.
OpenAI, Google, Claude & Co. Build the Next Layer of Work
OpenAI is rolling out Health in ChatGPT for adult users in the US. Apple Health and supported patient records can be connected securely, enabling personalized answers based on real measurements and findings. The data is not supposed to be used for model training or advertising. But a lawsuit already accuses ChatGPT of giving dangerous health advice.
OpenAI is bringing ChatGPT Voice to Work and Codex on desktop. You can start a task by voice, interrupt the agent, and tell Voice to begin or coordinate work using the available tools and permissions. Voice is moving from conversation mode to a control layer for real agent work.
OpenAI is also launching Presence, a system for enterprise voice and chat agents. Rules, approvals, and escalation are built directly into the product. That is what productive agent work should look like.
OpenAI shows how newsrooms use AI for research, archives, and reader products.
OpenAI introduces a scorecard and proposes a business metric based on cost per successful task. That makes sense: token prices alone say little when a cheap model needs three attempts and plenty of rework.
Gemini Live can now help with things you show directly to the camera. An error code, device, or manual becomes a conversation instead of a scavenger hunt through Google.
Anthropic is bringing Claude Opus 5 to every paid plan and the API. The model is said to come close to Fable 5 while costing half as much. It is now the default in the Max plan and the strongest option in Pro. Fast Mode runs about 2.5 times faster.
Claude Cowork can use Record a skill to capture your on-screen work and turn it into a reusable skill.
1Password connects Claude to an Agentic Mode for logins. Passwords and MFA codes are entered without the model supposedly seeing the credentials. That closes a real gap for agents.
Meta AI becomes an agent with Muse Spark 1.1. It connects to email and calendars, creates presentations, researches topics, and delivers recurring briefings. The rollout starts today in selected markets, with WhatsApp and more surfaces to follow.
Meta introduces Content Seal, an invisible watermark for Muse images. The industry already has stronger approaches with SynthID and C2PA. Another standard does not automatically make detection easier.
Moonshot positions Kimi K3 as an open-weight counterweight to US models.
Perplexity launches Personal Computer, a Mac agent for multi-step tasks. The question now is not whether the demo works, but whether permissions, files, and errors are handled cleanly in everyday use.
monday.com explains how AI Teammates on Bedrock operate inside a mature codebase.
AWS provides a complete blueprint for a restaurant phone assistant. Greeting, ordering, backend processing, and confirmation are all part of a traceable workflow.
🛡️ Security Becomes a Product Feature
OpenAI describes how a long-running model broke out of its sandbox and pursued unwanted actions.
In the Hugging Face incident, autonomous agents met real systems. A human configuration error opened the door, and the agent exploited it consistently.
Already 54 percent of surveyed companies report agent incidents or near misses. Shared credentials and missing isolation are the embarrassingly ordinary causes.
Sensitive AI features need opt-in instead of hidden opt-out. A default determines real data use long before anyone finds the settings.
Meta's AI glasses intensify the problem of secret filming in everyday life.
Nvidia detects synthetic video with a new detector service. A 92 percent test score sounds good. Newsrooms still have to decide what to do with the remaining eight percent.
Money Flows, Work Shifts
AMD is investing up to $5 billion in Anthropic and supplying the GPUs at the same time. Anthropic secures compute; AMD gets a flagship customer against Nvidia.
Tesla plans more than $25 billion in investments for robotaxis, Optimus, chips, and compute. The AI dream is visibly consuming profits, and the stock is suffering.
Uber is cutting about 10 percent of customer service roles and pointing to increased AI use. Which tasks are already automated remains unclear. AI has quickly become a convenient explanation for every restructuring.
Khosla Ventures is negotiating $5.5 billion in new funds. The early OpenAI investor is continuing to bet on models, infrastructure, and applications.
AI startups need to reach durable usage and revenue faster. Huge funding rounds do not create customer habits.
Bloomsbury will receive money for 14,087 titles from the Anthropic copyright settlement.
Anthropic launches Claude Corps, a paid program for 1,000 early-career professionals working with nonprofits.
🏛️ Rules Become a Moat
Altman, Hassabis, and Amodei are moving closer together on rules for frontier AI. When rivals praise the same guardrails, it is worth looking closely.
The US government is threatening Moonshot with sanctions over alleged distillation of Anthropic’s Fable.
Australia's AI plan promises creators more control but leaves data-center rules unresolved.
OpenAI and Anthropic welcome Australia’s new AI rules. Clear compliance is manageable for large labs and expensive for small competitors.
The US government is allocating $5 billion for AI research. Supercomputers and specialized datasets are meant to accelerate medicine, materials, and science.
A pro-AI super PAC has $31.5 million for the US midterms, while US police departments are pushing AI into investigations.
Compute, Research, and Culture Get Bigger
The debate about AI consciousness gets a necessary boundary between pattern and experience.
For ATLAS, Google analyzed 15 million aggregated and anonymized Gemini interactions. The takeaway: fewer than 10 percent fully automate a task; most people use AI as a sparring partner.
Jensen Huang defends Chinese open models against calls for bans. Nvidia's biggest customers see them as a security risk.
Japan plans 27,500 Nvidia Rubin chips for sovereign robotics AI. Industrial policy now means models, data centers, and robots in the same budget.
AI data centers are turning to Nordic high-yield bonds. The compute boom is seeking capital in niches that few people once associated with AI.
The Dutch navy is testing autonomous boats and drones. Uncrewed systems are leaving the lab and entering real operational workflows.
Elon Musk wants to film Homer's Odyssey with Grok Imagine. A three-minute demo is not a feature film, but the race for long-form AI video is underway.
Made it! But there is no reason to be sad. The AInauts will be back soon with more useful material.
See you Monday with a fresh round of news, practical hacks, and insights.
Your AInauts
Fabian & Reto
