In partnership with
Welcome, AI enthusiasts
Anthropic's AI now writes over 80% of the company's code, one of its engineers hasn't typed a line in five months, and its policy chief thinks AI could start building its own successors within two years. So this week the lab did something no frontrunner has done before. It published the numbers and asked for a way to slow itself down. Let's dive in!
In today’s insights:
AI's Point of No Return Is Near
AI built a vaccine to STOP the next pandemic
ChatGPT Memory Becomes the New Lock-In
Read time: 3.5 minutes
LATEST DEVELOPMENTS
ANTHROPIC
🚨 AI's Point of No Return Is Near
Evolving AI: Anthropic warns AI may soon improve itself beyond human oversight.
Key Points:
Recursive self-improvement could be two years away according to Anthropic's policy chief.
Human judgment over what to build next is the last clear advantage left.
The company wants systems that could verify a global pause.
Details:
Recursive self-improvement means an AI designing its own successor with no human in the loop. Anthropic's evidence comes from its own workflow where Claude now writes over 80% of merged code and one engineer hasn't typed a line himself in five months. Choosing which problems to pursue remains the human edge, and the company's own measurements show that gap closing faster than expected.
Why It Matters:
That shrinking gap is the reason Anthropic wants to act now. The company is asking for the option to slow down before the loop closes, but no system exists to verify a rival lab actually paused. Training runs are far easier to hide than missile silos, which is why Anthropic says it will start building those tools itself.
Claude is not just a chatbot anymore. Is your security team ready?
Claude.ai is one thing. Agentic workflows, MCP connections, ungoverned skills taking actions across your data? That's a different conversation — and most security teams aren't equipped for it.
Harmonic Security gives your CISO the visibility and controls to say yes confidently.
HEALTH AI
💉 AI built a vaccine to STOP the next pandemic
Source: Science Source
Evolving AI: University of Cambridge team put an AI-designed vaccine into human trials.
Key Points:
The core of the vaccine was designed entirely by AI, a first for the technology in humans.
The shot targets the whole coronavirus family, including strains still circulating in animals.
Safety testing covered 39 people, with a study of around 200 now underway.
Details:
Researchers fed the AI genetic data from many coronaviruses flagged by surveillance programmes, instead of the usual approach of building around today's virus. From that, the system designed a super-antigen. The idea is to prime the immune system against a whole family of related viruses, including ones that mutate or jump from animals. Results in the Journal of Infection showed a modest immune response, though the team reads it as an early signal, not a finished result.
Why It Matters:
Vaccine work has always trailed fast-moving viruses, and that lag is the whole headache behind flu and Covid boosters. AI has been moving steadily into drug design, and aiming it at threats that have not emerged yet could shift preparation earlier. The response was soft, so the team is already trying the approach on flu, Ebola and bird flu to see how far it travels.
The best prompt engineers aren't typing. They're talking.
Power users figured this out early: speaking a prompt gives you 10x more context in half the time. You include the edge cases, the examples, the tone you want — because talking is fast enough that you don't skip them.
Wispr Flow captures everything you say and turns it into clean, structured text for any AI tool. Speak messy. Get polished input. Paste into ChatGPT, Claude, Cursor, or wherever you work.
89% of messages sent with zero edits. 4x faster than typing. Works system-wide on Mac, Windows, and iPhone.
Evolving AI: OpenAI's new Dreaming system keeps ChatGPT's memory of you fresh.
Key Points:
Dreaming synthesizes memories from chat history automatically in the background.
Entries revise themselves as time passes instead of going stale.
Compute costs dropped roughly 5x, with free accounts next in line.
Details:
Dreaming replaces the old approach where ChatGPT only kept what users explicitly asked it to remember. The system now runs in the background and picks up context that comes up naturally in conversation. It also rewrites outdated entries so the model stops thinking you are still on a trip that ended months ago. Plus and Pro users in the US get it first with free accounts following over the coming weeks.
Why It Matters:
Extending memory to free users is where this turns strategic. Anthropic and Google spent this year building import tools that let people carry memories between chatbots. People tend to stay with the assistant that already knows them, and ChatGPT is about to know millions more.
Most people still think prompting is about writing better instructions. Anthropic's Prompting Playbook argues the real advantage now comes from designing AI workflows that can plan, verify their work, and use tools when needed. It's a practical look at how prompting is evolving into a new discipline for building reliable AI agents.
👀 Click on the image you think is real
QUICK HITS
☁️ Cloudflare and xAI Make Grok a Native AI Gateway Model.
🤖 Nvidia's New Nemotron 3 Ultra Targets Agents That Plan, Reason, and Stick the Workflow.
💰 Devin Maker Cognition Will Cover Your Usage If the AI Doesn't Earn Its Keep.
🌊 Apoha Lands $36M to Design Proteins, Paints, and Foods With Physics-Based AI.
📈 Trending AI Tools
📝 Granola - AI notetaker that captures the real insights and turns every conversation into ready-to-share, action-driving notes*
💬 HelpCrunch - AI-powered customer support, live chat, and email marketing.
🎥 Pictory - Turns scripts, blogs, and long videos into short branded clips.
🐳 DeepSeek - Open-source AI models and free chat assistant.
*partner link