In partnership with
AI-HOY and happy weekend, AInauts!
Missed some of last week's AI news, tools and hacks, or just joined us? No worries. Here's our recap of the highlights:
We finish with the week's Quick News, so you can catch up on the developments that matter in one place. Ready? Let's go!
Your meetings, inside Claude and ChatGPT
Last week's customer call has the details your proposal needs. Right now they're buried in notes nobody can find.
Wispr Flow Notetaker captures your meetings with names, numbers, and product terms spelled right, then brings them into Claude or ChatGPT through a built-in connector. Ask Claude to draft the proposal from your last three calls with a prospect, or pull together what the deal team said about pricing this quarter.
Need something from a call in March? Ask one question across every meeting you've had and get the answer with citations. Get Wispr Flow Notetaker, free to try on Mac and Windows.
The same AI that saves us work is saving attackers work too. Critical PHP flaws, a breached cloud host and government portals taken offline across Europe made that hard to ignore this week. Nobody can be completely secure, but you do not have to be the easiest target.
Our Deep Dive tackles email, 2FA, phones, updates, backups and routers in order, with a separate section for website owners and agent builders.
Short on time? Use the Security Starter Kit: your agent checks your setup and tells you what needs attention. Nothing changes without your approval.
Got an agent but no job for it? Another model update won't help much. Start with your most annoying routine, then choose the tool. Language note: the following directory and its prompts are in German.
usecases.ainauten.com now has 345+ practical use cases with prerequisites, steps and prompts. One of our agents maintains it. Of course. Pick a task that fits your daily work and try it this week.
State of Product reveals what AI still hasn’t solved
80% of product professionals say AI helps them ship faster, but customers aren’t seeing value any sooner. Atlassian’s State of Product 2027 explores the gains AI is delivering and the challenges that remain, from decision-making that hasn’t kept pace to gut instinct overriding customer evidence.
A browser agent clicks through free courses and solves the final tests. We tried it and came away with 62 more certificates. The freely available instructions even tell the agent to continue where providers prohibit AI help on exams.
The badges are less interesting than the skill: if you can get an agent to finish reliably, you can hand it quotes, research or your weekly report. A Bitkom survey shows the gap: 57% of German companies use AI, but only 3% of users say they make substantial use of its potential.
Your research sits in ten tabs, but the slides are blank. We closed that gap with ChatGPT's browser extension and the new ChatGPT Slides: upload your deck as a template, have ChatGPT analyze the tabs, and build the presentation.
Edit the result directly and export to PowerPoint, Google Slides or PDF. Public sharing wasn't available in our test, so we turned the slides into a website. Slides initially requires Pro, Business or Enterprise with Space; the extension is for everyone. That puts pressure on Gamma when a quick meeting deck comes included in chat.
One developer had Claude coordinate eight agents in parallel to build open-source counterparts to Photoshop, Illustrator and more. They're far from finished, but the Photoshop clone already opens PSD files with layers and masks.
Our community's examples are closer to daily life: tracking private health insurance receipts, dictating construction-site defects and extending old Access databases. We rebuilt Android's missing 360-degree panorama feature ourselves. Our rule: build when you need one feature or a fraction of an expensive subscription; keep buying when others depend on it or sensitive data is involved.
"Can I turn off AI training and then upload customer data?" It isn't that simple. We compared providers with different approaches and built a privacy ladder: from free chat through business plans and EU platforms to local processing.
The right level depends on your data, not the nicest seal. For lawyers, tax advisers and doctors in Germany, a data processing agreement alone isn't enough. This isn't legal advice, but it gives you questions to ask your provider, plus two free helpers that replace names and IBANs before they reach the chat.
An internal memo says Accenture centrally covers only $29 per person per month for Claude; a project must fund anything beyond that. The question is reaching every team: what does AI deliver? In HubSpot's survey of more than 2,800 marketing leaders, 83% said they're under pressure to prove ROI.
We show five ways to measure impact, including an honest baseline, review and waiting time, and cost per result rather than per month. Pick one recurring process and measure before the next license renewal.
Language note: HubSpot's free masterclass on November 4 is in German and serves the German-speaking market.
You close the laptop and want your agent to keep working. Sometimes it does; sometimes that's the end of its shift. We separate three approaches: agents with their own cloud computer, such as Dots or Grok Bot; scheduled tasks in ChatGPT and Claude; and remote access to a computer you've left on.
Where your data lives is decisive. A cloud task can access Gmail, calendars and Drive, but not your Downloads folder. We've included a ready-to-use prompt for a morning briefing from your inbox, so you can try it.
AI News Quickie: This Week's Highlights
AI never sleeps!
This week, OpenAI and Anthropic both move into Google Cloud, Gemini becomes an office assistant for small businesses, and Canva and Ideogram get their own image models. Meanwhile, PHP, Railway and OpenCode show why your access permissions and keys deserve more attention than the next feature.
💬 OpenAI: Lock Chats and Handle Small Tasks Faster
Locked chats disappear from normal history and search and open only with a PIN or device unlock. The gradual rollout includes the free plan. This is a screen lock, not encryption: storage and training settings stay unchanged.
ChatGPT Ads is now available in 89 countries. More than 100,000 advertisers have run campaigns since launch. Ads appear only on Free and Go; Plus and Pro remain ad-free.
Want the technical details? OpenAI explains how its ad system operates separately from answers.
OpenAI models and Codex are coming to Google Cloud, with Google's access controls and billing through the Marketplace. General availability is expected in the coming weeks.
The Decisions API is available: GPT-6 Luna makes small decisions such as 'invoice or support?' from predefined answers. OpenAI says it is up to ten times faster than the standard API and supports up to 400 questions per request.
If you run your own AI applications, you can also cache context for twelve hours.
🔎 Google and Claude: Office Help Instead of More Chat Windows
Gemini is getting a business area for small companies: save brand materials, create matching campaign images, analyze receipts and spreadsheets, and connect Canva, Shopify or Xero. The global rollout starts in the coming days for all Gemini app users.
In Gemini Notebook, you can check and change slide outlines and explainer-video sequences before rendering. It arrives this week. Finally, fix the plan before having to curse the finished video.
The full Claude Platform is available on Google Cloud, with Google billing. Anthropic operates the service and processes the data, currently through US and global endpoints. A Google invoice doesn't automatically mean EU processing.
Anthropic is handing the open Agent Skills standard to the Agentic AI Foundation under the Linux Foundation. More than 70 agent platforms already support it, including ChatGPT and GitHub Copilot. Your skills can work across tools rather than in just one chat.
Claude Code now warns when resuming old conversations if rereading them would noticeably consume the five-hour Pro or Max allowance. You can continue or start fresh and reference the old conversation.
Claude Security scans code continuously and tracks whether vulnerabilities were actually fixed. It's a public beta for Enterprise customers, not a free check included in every Claude subscription.
Anthropic plans to put at least 5% of its shares into a Public Benefit Fund, subject to shareholder approval. The fund would address issues such as effects on the labor market. More to come.
🇪🇺 Europe: Nvidia Backs Germany, Mistral Backs Voice
At the Berlin investment summit, Germany's government and Nvidia signed a memorandum on industrial AI.
Nvidia also plans to invest in four German startups, including Agile Robots and Avelios Medical.
At Siemens in Erlangen, agent-based production planning is expected to cut costs by up to 20%.
Nvidia's RTX Spark laptops from Asus, Dell, HP, Lenovo, Microsoft and MSI are on sale, with up to 128 GB of unified memory for local models, starting at $2,399. Nvidia gives no euro prices. Worth watching if your data shouldn't leave your computer.
Mistral's Voxtral 3 combines listening, responding and actions in business systems into voice agents, with transcription in more than 50 languages. Use the EU-hosted API or self-host it.
🧰 Agents and Models: Hand Off More Work
ElevenLabs is gradually rolling out Eleven Realtime, a new model family for voice agents. Instead of combining a language model, speech recognition and speech synthesis yourself, choose a complete configuration.
DeepSeek V4.1 Flash 1017 is available on the website, app and API and aims to finish coding tasks faster with shorter outputs. The weights are already downloadable.
Perplexity Computer now runs fully locally on Macs. It requires a supported Apple Silicon Mac; at least 36 GB of memory is recommended. Cloud processing is used only if you allow it.
🎨 Images and Music: Original Models, Less Mass-Produced Content
Canva Craft, Canva's first proprietary image model, is available worldwide in Dream Lab and through the API. It plans layout and text before generating the image, then Magic Layers can separate it into editable elements: material you can work with rather than a finished block of pixels.
Ideogram has launched Ideogram Vector, its first vector model. Text prompts become editable SVGs, and existing images can be converted to scalable vectors, directly in Ideogram or through its API. It gives you a starting point for logos, icons or print graphics in Illustrator or Figma; the finishing work is still yours.
Spotify has removed more than 50 million slop and spam tracks from recommendations this year, on top of more than 75 million spam tracks removed from the platform. Thousands of AI personas now carry a label and are excluded from recommendations by default unless you follow them.
ElevenLabs also has a Video-to-Music API: send a video and receive a matching soundtrack without a music prompt. Optional sound effects and separate music/effects tracks let you mix it yourself. For product videos, ads or social clips, it's a route to a first audio track, initially API-only.
🛡️ Security: Permissions You Should Actually Check
Critical PHP vulnerabilities, affecting technology behind WordPress and many other sites, were found with AI's help. Updates have been available since October 21, but updating a WordPress plugin isn't enough. Ask your host whether your site uses the patched PHP version.
Credentials leaked at Railway, including API keys and database credentials from stored environment variables. All users should rotate their API keys.
Attackers stole stored API keys from coding agent OpenCode for OpenAI, Anthropic and Google and used them at users' expense. If you use API keys, set a spending limit with the provider.
Shared Muse conversations were publicly discoverable because titles and link identifiers appeared in searchable DNS data. Meta has suspended the feature. Don't share confidential chats through public links: a long random address isn't access control.
Zoom announced warnings about possible video and audio deepfakes and centralized controls for AI agents. Both are planned for December.
GitHub's new package firewall for Enterprise Cloud can block suspicious software components before installation, including those a coding agent tries to download.
🔬 Research and Medicine: Progress Is Not a Finished Product
Microsoft's diagnostic system MAI-DxO was correct in 80% of emergency-department cases in an evaluation now published in Science, versus 47% for doctors. But a benchmark doesn't treat a patient.
An internal Claude research model designed protein binders against scorpion toxins. Of 211 designs tested in the lab, 17 met the binding criteria. That isn't an approved antivenom yet.
OpenAI, PerturbAI and Scripps deployed thousands of agents on mouse-brain data. The analysis and literature review took 32 minutes instead of months. The results offer hypotheses for new experiments involving Parkinson's or Alzheimer's therapies.
OpenAI has acquired Orchestra Bio, a platform coordinating drug development. Agents will track studies, dependencies and next steps: even in the lab, there's plenty of admin between insight and outcome.
That's it for this week. We'll be back soon with more news, hacks and insights.
See you Monday!
Your AInauten
Fabian & Reto
P.S. Our social channels are primarily in German: X, LinkedIn, Facebook, Instagram, YouTube and TikTok.
