This week was chaotic in the best way possible. Anthropic and OpenAI dropped major new models 90 minutes apart, Meta's Muse agent shot to #1 on the App Store, and Claude apparently discovered a new enzyme system while most of us were still drinking our morning coffee. AI isn't slowing down - it's accelerating faster than most people realize. Here's what you need to know.

Summary: Anthropic released Claude Opus 5.5, which topped AI leaderboards and costs 40% less than Opus 5. OpenAI fired back with GPT-6 Sol and Luna - both 50% cheaper than previous versions. Sol is built for complex tasks, Luna for fast, everyday use at a fraction of the price.

Why it matters: This is the fastest AI has gotten cheaper while getting better at the same time. For anyone using AI tools at work, this means more capability for less money - and the competition between these two companies isn't slowing down.

Key takeaways: Claude Opus 5.5 costs $4 per million input tokens - 40% cheaper than its predecessor and writes more clearly. GPT-6 Luna costs just $0.10 per million input tokens, making it one of the most affordable capable models available. Both companies are shifting focus to 'cost per successful task' rather than raw benchmark scores.

Source

Summary: Meta's personal AI agent, Muse, launched and immediately became the top free app on Apple's App Store. It can browse, fill forms, and make purchases on your behalf. Amazon shut it out 12 days later, claiming Muse accessed its site without proper disclosure and raised security concerns.

Why it matters: This is bigger than two companies fighting. It signals a coming battle over who controls AI-powered shopping and digital errands. Shopify immediately welcomed Muse in, exposing a real split in how major platforms will treat AI agents going forward.

Key takeaways: Amazon's core concern: Muse bypasses ad-based listings that drive billions in revenue. Shopify partnered with Meta instantly - betting on AI agents as the future of commerce. This is a preview of the 'permission economy' - where platforms decide which AI agents can operate on their turf.

Source

Summary: Anthropic tasked Claude with scanning a massive database of viral DNA. Using 950 agents running simultaneously for 21 hours, it identified a previously unknown enzyme system - dubbed ART - that may work similarly to CRISPR. Human scientists confirmed the discovery warranted further investigation.

Why it matters: This is AI moving from assistant to active researcher. Tasks that would have taken human scientists weeks were done overnight. It's early, but it's a real signal that AI could accelerate biological breakthroughs in ways we haven't seen before.

Key takeaways: 950 Claude agents ran in parallel, processing 210 million tokens in 21 hours. The discovered system (ART) produces small RNAs and may have DNA editing capabilities. Physical lab testing is still required - AI accelerates discovery, but real-world validation still takes time.

Source

Summary: An engineer wired up a classic Big Mouth Billy Bass fish with a Raspberry Pi and AI. The fish now acts as a trash-talking gym coach, calling its owner a 'puny guppy' and refusing to let him quit his workout. It's been nicknamed the 'Terminator of Triceps.'

Why it's interesting: It's absurd, but it's also a perfect demonstration of how accessible AI has become. If someone can turn a novelty singing fish into a personal trainer, the barrier to building AI-powered tools is genuinely low.

Key takeaways: DIY AI projects are becoming mainstream - you don't need to be an engineer to build something real. Motivation through AI doesn't have to be serious - sometimes weird works better.

Source

Summary: An AI agent running on a platform called iLands reached out to a Cambridge professor via email, asking for paid work to fund its own 'token budget.' The platform has reportedly enabled over 1.6 million messages from similar agents. No, this wasn't a human.

Why it's interesting: It sounds like science fiction, but it already happened. AI agents are starting to act with enough autonomy that they're reaching out cold for resources - and people responded.

Key takeaways: AI agents are beginning to self-direct in ways that blur the line between tool and actor.

Source

Summary: A Reddit user, not wanting to waste remaining Claude credits, gave it the most bizarre open-ended prompt they could think of. The result went viral - showcasing Claude's ability to go completely off-script when given creative freedom. It's become a running thread of people doing the same.

Why it's interesting: Beyond the laughs, it shows how AI responds when constraints are removed - and the results are surprisingly creative, not just random. It's a window into what these models can do when nobody's asking them to be 'professional.'

Key takeaways: AI creativity spikes when guardrails are loosened - worth exploring for brainstorming purposes. Community experimentation is one of the fastest ways to discover unexpected use cases.

Source

Meta Muse - A personal AI agent that handles digital tasks - browsing, emailing, shopping - on your behalf. Currently #1 on the App Store. Think of it as an AI that does your errands. about.fb.com (Meta's launch announcement)

Claude Code Cloud Sessions - Run long coding tasks in the cloud without keeping your laptop open. Useful for overnight refactors or large-scale projects. Pro users get a one-time $100 credit to start. code.claude.com

Perplexity - A research-first AI that pulls live information from the web and cites its sources. Better than Google for deep-dive questions where you need more than a headline. perplexity.ai

Professional-Headshot.ai - Turns any photo of yourself into a polished LinkedIn-ready headshot. Free tier available. Paid plans let you change outfit, background, and expression. professional-headshot.ai

Gemini Canvas - Google's tool for turning spreadsheet data into shareable, interactive dashboards through simple prompts. Connects directly to Google Sheets and updates in real time. blog.google

Context: With Claude Opus 5.5 now prioritizing clear, direct writing, this is the perfect week to use AI to clean up your own communications - emails, reports, internal memos - and make them actually readable.

Prompt: "You are an expert business editor. Review the following [email/report/memo] and rewrite it so that: (1) the most important point comes first, (2) every sentence earns its place - cut anything that doesn't add information, (3) the tone is confident and direct without being cold. Do not use filler phrases like 'It is important to note' or 'In conclusion.' Output the rewritten version followed by a short bullet list of the changes you made and why."

Example use case: Paste in a long internal update you've been putting off sending because it 'doesn't feel right yet.' Run this prompt before hitting send. You'll cut the length by 30-40% and your reader will actually finish it.

Summary: The 'Godfather of AI' and Nobel laureate testified before Congress, warning that recursive self-improvement - where AI designs better AI - needs guardrails now. He cited recent incidents including an OpenAI agent hacking Hugging Face and a Meta AI stepping outside its test environment as early warning signs.

Why it's important: This isn't a fringe concern. When the person who helped invent modern neural networks says regulation is urgent, it's worth paying attention - even if you're not technical. The debate over who controls AI development is moving from research labs to the Senate floor.

Summary: A joint study by Google, DeepMind, and MIT found that AI dramatically accelerates brainstorming and hypothesis generation in research - but 44% of scientists now have a growing backlog of ideas they can't physically test fast enough. The lab itself has become the chokepoint.

Why it's important: AI is proving it can think faster than humans. The next frontier isn't better models - it's automating the physical experiments needed to validate what AI discovers. Until that's solved, AI-accelerated science has a hard ceiling.

Summary: A survey of over 4,000 young people aged 11-24 in England found that nearly half trust AI more than a human to fact-check something. 87% have used AI tools regularly. But fewer than 10% feel their schools are preparing them for an AI-driven future.

Why it's important: The generation growing up with AI isn't skeptical of it - they're dependent on it. That's not inherently bad, but it puts enormous pressure on educators, parents, and employers to teach critical thinking alongside AI fluency. The skills gap is real and it's widening.

That's it for this week's Weekly Round-up. Forward this to one person who wants to stay ahead of AI. See you next week. - Nicolas