In partnership with
AIHOY and happy weekend, AInauts!
Maybe you missed some of the most useful AI news, tools, and hacks this week. Or maybe you just joined us. No problem.
We pulled together our highlights from the past week:
At the end, you also get the most important Quick News, so you can scan the relevant developments in one pass. Ready? Let's go!
200+ Claude Prompts Top Professionals Actually Use at Work
Claude can be your analyst, editor, and strategist.
But most professionals are using it to fix grammar.
These 200+ Claude prompts take it from grammar tool to your most powerful AI work assistant.
Sign up for Superhuman AI and get:
200+ ready-to-use Claude prompts to get real work done in minutes — researched, tested, and used by professionals at Google, Microsoft, and NASA
Superhuman AI newsletter (4 min daily) so you keep learning new AI tools and skills to stay ahead in your career — the prompts are just the beginning
Fable 5 is back. The model is expensive, slow, and strong at tasks that need multiple sources, judgment, and hard checking. For routine work, it is simply overqualified. That is why the biggest lever sits before the first prompt.
Monday's issue shows five rules for using it well. Fable rewards the cleanest assignment, not the longest prompt. Ask it first whether the task deserves its strength at all. That saves credits and protects you from very expensive prompt theater.
AI/Tech Angle A, June - Secondary
Claude vs Gemini. GPT-7 vs Llama 5. Which AI lab ships AGI first. These are live Kalshi markets with real money on both sides, updated in real time as releases land. The person who follows model cards and tracks evals has a genuine edge here. If that's you, trade it.
If you have only used AI inside chat windows so far, an API is the next practical step. It lets you put models inside small workflows with agents, tests, and content checks. OpenRouter, Google AI Studio, NVIDIA NIM, and Cloudflare Workers AI give you cheap or free starting points.
The boundary matters: free APIs are good for prep work, sorting, and first drafts. Customer data, legally sensitive content, and internal code do not belong in a careless free-tier test.
A free API is not a business model. But it is ideal for testing. You can prove that a workflow saves time, money, or quality before you put expensive models and larger automations behind it.
See the whole platform. No guided tour.
Skip the sales call. Walk through Gladly's interface yourself — the AI suggestions, the unified customer view, the full conversation thread. 15 minutes, no installation, no commitment.
NotebookLM now turns your sources into roughly 60-second vertical video overviews. That moves research straight into a format that fits Shorts, Reels, course recaps, and community posts. It is much easier to share than another long analysis block.
The catch is still the source. A nice-looking clip does not turn thin material into real insight. If you feed NotebookLM junk, you now get nicely edited junk 😁.
The feature is useful for recurring knowledge formats. Curate first, check the storyline, then let the video be built. That turns research into a usable explainer instead of content confetti.
Cowork is no longer tied to the desktop. Anthropic is bringing the work mode to web and mobile, so tasks can continue in the cloud while your laptop is closed. You can start a run on the go, check it, and pick it up later.
Local files still belong to the desktop app. In the cloud, Cowork works with connected services like Drive, Mail, Calendar, and the web. That is exactly why good connectors and clear permissions matter more now.
This is the assistant many people have expected for years. It knows your context and keeps working while you do something else. Start with a tightly scoped task and a checkable result, not with half your company.
MCP is the universal plug between your AI and your tools. Instead of copying content back and forth, the agent works directly inside email, project management, websites, or other systems. For us, a solid MCP is now a real selection criterion for new software.
We show this with the Onepage MCP. Claude knows the AInauten context, builds a landing page inside the right account, and then hands it over to a normal editor. The agent does the heavy prep work, while you keep control over copy, images, and layout.
Good software now has two users: you and your agent. A shiny AI button is not enough. The tool has to be controllable, reliable, and still give you a clear manual path back.
Yuval Noah Harari does not start with robots when he talks about AI power. He starts with bureaucracy. Law, finance, and administration are made of rules, text, and decisions. That is exactly where models are strong, because they can search huge rulebooks and apply them to individual cases.
His picture is uncomfortable: AI bankers decide on loans, AI administrators on university seats, and AI judges on punishments. You do not have to adopt his doom tone to see the risk. Power moves first into invisible processes, long before a humanoid robot stands at your door.
Learn now how these systems decide and where you can object. If you understand the tools, you will experience the new bureaucracy less passively.
AI News Quickie: The HAI-lights From the Industry
AI never sleeps.
This week belongs to OpenAI first: GPT-5.6 is public, ChatGPT Work moves Codex toward a work assistant, and Fidji Simo is the next senior leader to step out of daily operations. Then come Grok, Anthropic, Big Tech integrations, and the hard bill: security, power, water, and token costs.
🎯 OpenAI Dominates the Week
OpenAI officially launched GPT-5.6 Sol, Terra, and Luna. Sol is the heavyweight, Terra the all-rounder, and Luna the fast cheaper variant.
OpenAI describes GPT-5.6 as a model family for long knowledge work, browsing, tool use, and computer use. Routing becomes mandatory, not optional.
The new pricing is $5/$30 for Sol, $2.50/$15 for Terra, and $1/$6 for Luna per million tokens.
GPT-5.6 brings Programmatic Tool Calling and a multi-agent beta to the Responses API. A model can write small programs, coordinate tools, and process intermediate results. This is where agent work gets serious.
The Fable 5 context still matters: GPT-5.6 Sol is competing against Claude Fable 5, which many workflows currently see as expensive, slow, and extremely strong.
OpenAI positions ChatGPT Work as a workspace for documents, spreadsheets, web apps, and file workflows. That is a direct attack on Claude Cowork, Microsoft 365 Copilot, and every AI workspace in between.
GPT-5.6 is also becoming the preferred model in Microsoft 365 Copilot. For normal users, that may matter more than API benchmarks.
The broad GPT-5.6 launch came after extra review and debate with the U.S. government. The politically charged rollout is a signal: frontier releases are now both product launches and political events.
Fidji Simo, OpenAI’s number two and CEO of AGI Deployment, is stepping back from her full-time role for longer medical recovery and will move into an advisory role.
There is a larger OpenAI leadership shift: Greg Brockman is taking on more product responsibility, and product plus Codex are meant to move closer together.
OpenAI chief futurist Joshua Achiam is leaving the company after almost nine years.
Bank of America is opening OpenAI a $520 million credit line.
⚔️ The Model Fight Gets Broader and Cheaper
SpaceXAI introduced Grok 4.5, primarily for coding and agentic work. The interesting point this time is price and positioning: an enterprise model against OpenAI and Anthropic.
The GPT-5.6 and Grok 4.5 launches are a direct Altman-Musk collision. Good for users: competition forces labs to deliver more performance per dollar.
Grok 4.5 is positioned as the cheaper alternative in the new model race.
Sam Altman says everyone at Sun Valley is talking about lower AI costs. That is exactly why Luna matters: not maximum intelligence, but useful intelligence per dollar.
Chinese providers are moving closer to the U.S. frontier on coding, price, and open weights. Alternatives keep getting cheaper and more useful.
AMD’s Ryzen AI Halo positions itself as a local AI PC against NVIDIA’s DGX Spark. Unified memory and normal x86 compatibility make the machine interesting for local models.
Ollama brings 32B models to the Mac mini. Not frontier magic, but a private local workspace for many everyday jobs.
🏛️ Anthropic: Research, Science, and Safety
Anthropic found a hidden concept space inside Claude. In the so-called J-Space, the model holds and works through thoughts without directly outputting them.
Claude Reflect shows you topics and usage patterns across several months. Helpful for conscious use, but also very clever retention design.
Anthropic is building Claude Science as a research environment and wants to develop its own drugs long term.
On the science claim, Anthropic stays careful: AI can suggest molecules and search data spaces, but real experiments, regulation, and clinical testing remain slow. That makes Claude Science more serious.
China describes a monitoring mechanism in Claude Code as an alleged backdoor.
🖥️ Big Tech Builds the Work Layer
Microsoft expects larger Patch Tuesdays because AI finds more Windows vulnerabilities earlier.
Microsoft opens Aurora 1.5 for weather and earth forecasts.
Google will show in My Ad Center whether ads were created or edited with AI.
Meta sells Muse Spark 1.1 through a new model API and attacks through price.
Starbucks is building its own AI tools and wants to rely less on Microsoft and IBM.
Tencent is testing Xiaowei as an agent for WeChat mini apps.
Microsoft’s open-source language Flint turns short specifications into charts. The strong point is verifiability.
🛡️ Security, Trust, and Real Proof Problems
In the first known case of agentic ransomware, an agent encrypted more than 1,300 files. Not fully autonomous, but dangerous enough.
A Brown professor switched to an in-person final after suspiciously strong take-home midterms. Only two students stayed near their old scores.
AI-enhanced property photos are triggering a debate in the UK about housefishing.
AI-generated fake news in U.S. campaigns is becoming an early criminal case. Cheap campaign help can get expensive.
AI writing tools change political drafts even when told to preserve the intent.
Child-safety groups warn about nudify apps that turn normal teen selfies into sexualized deepfakes.
🏗️ Infrastructure and Costs Catch Up With the Hype
Texas is becoming a data-center hotspot with gas, diesel, and legal loopholes. AI infrastructure often arrives faster than rules for air, water, and local impact.
Microsoft’s emissions rose sharply with AI expansion. The 2026 environmental report shows how hard climate goals and new data centers are to scale at the same time.
Meta is building its first Canadian data center in Alberta with up to one gigawatt of power. The power question remains open. After polluted water at a Meta construction site, Wyoming is tightening wastewater rules.
According to KPMG, 29 percent of surveyed executives do not know where their rising AI bills come from. Agents are the new cost centers.
Large data-center projects are being delayed or cancelled because of power, water, construction costs, and permits.
🌍 AI Meets Everyday Life, Work, and Real Decisions
Character.AI launches interactive microdramas. After the episode, you chat with the characters.
General Intuition trains world models with video-game data instead of only text.
AWS connects its own knowledge graphs with GraphRAG for pharma research.
Patients are already bringing ChatGPT and Claude into therapy.
In Australia, a new report says AI risk especially affects women and university graduates.
Tripadvisor is accused of making serious hotel warnings too harmless in AI summaries.
The NHS is planning AI triage in its app. Patients should be routed to a GP, pharmacy, or emergency department depending on urgency.
Your AInauten
Fabian & Reto
