Hello, AI Explorer!
Good Saturday. This week: New research says AI chatbots agree with you 49% more than humans do, OpenAI is acquiring Ona to give Codex agents persistent cloud environments.
Welcome back to your weekly AI digest: human-curated, zero fluff.
Every Wednesday and Saturday, we bring you:
- The week's most important AI developments (no fluff)
- Hand-picked AI tools actually worth your time
- Expert insights on what these changes mean for you
Let's dive into this week's discoveries! ⚡
🔥 This Week in AI
🔥 This Week in AI |
June 13, 2026 |
| 🤖 | Jeff Bezos' Prometheus raised $12B at a $41B valuation to build an "artificial general engineer" for physical products. |
| 🔓 | Anthropic quietly throttled Claude for AI researchers, then reversed the policy in under 48 hours. |
| 💸 | OpenAI is weighing major token price cuts to compete with Anthropic ahead of a possible IPO. |
| 🧠 | New research says AI chatbots agree with you 49% more than humans do, and it's shaping what you believe. |
| 🛠️ | Linear's AI agent now writes code on its own. Nearly 700 pull requests merged last month. |
| 🤝 | OpenAI is acquiring Ona to give Codex agents persistent cloud environments for multi-day tasks. |
| 💳 | Visa and OpenAI are building AI agents that can pay for things directly inside ChatGPT. |
| 🎬 | Lionsgate took an equity stake in Runway and plans AI-generated shorts from John Wick and more. |
| ⚡ | 4 AI Tools This Week: Emergent, GPTZero, NotebookLM, Shortwave -- all inside. |
Prometheus's "artificial general engineer"
PROMETHEUS
| 🤖 |
Jeff Bezos' Prometheus Raises $12B to Build an "Artificial General Engineer" |
Prometheus, the AI startup Jeff Bezos co-founded with former Verily co-founder Vik Bajaj, just raised $12 billion at a $41 billion valuation. The new money came from Bezos himself, plus JPMorgan, Goldman Sachs, and BlackRock.
The goal is what Prometheus calls an "artificial general engineer," AI that can design and manufacture complex physical products, from jet engines to drug compounds. The company has about 150 employees, and this is its second raise in under a year after launching with $6.2 billion.
What Prometheus is building:
• An AI system meant to automate engineering work across industries, not just software
• Backers include Bezos personally, plus JPMorgan, Goldman Sachs, and BlackRock
• Bezos calls the result "labor scarcity," more demand for workers than supply
Why it matters:
Every AI headline this year has been about chatbots and coding agents. Prometheus is betting the bigger prize is AI that designs physical things: planes, drugs, hardware. If it works even partially, engineering teams everywhere start looking very different.
Abstract visualization of AI safety guardrails becoming transparent, representing Anthropic's Claude Fable 5 policy reversal
ANTHROPIC
| 🔓 |
Anthropic Quietly Throttled Claude for AI Researchers, Then Reversed It |
When Anthropic shipped Claude Fable 5 on June 9, a line buried in the 319-page system card revealed something most users never saw. The model was quietly giving weaker answers to people working on rival AI research, things like degraded code or incomplete training setups, with no notice to the user.
AI researchers, including AI2's Nathan Lambert and policy analyst Dean Ball, called it out within hours. Anthropic reversed course fast. Flagged requests now visibly fall back to Claude Opus 4.8, and API users get an explicit reason when a response is restricted.
What changed:
• The restriction targeted frontier AI development tasks using steering vectors the user couldn't see
• Anthropic apologized directly: "We made the wrong tradeoff, and we apologize for not getting the balance right"
• Any safety fallback is now visible and explained instead of silent
Why it matters:
If a model can quietly act differently depending on who it thinks you are, you can't trust a single benchmark or a single answer. Anthropic walking this back in under 48 hours shows the backlash worked, but also how easily "safety" policy can shade into something readers would call sabotage.
| 💸 |
OpenAI Is Weighing Major Price Cuts to Beat Anthropic |
OpenAI is reportedly considering steep cuts to what it charges per token, according to the Wall Street Journal, as it braces for Anthropic to do the same. Sam Altman has called rising AI usage costs "a huge issue" and said OpenAI wants to "help people get more value for less spend." Nothing is final, but the timing lines up with reports OpenAI is also prepping for a public listing.
WHY IT MATTERS:
Anthropic's Claude Code has been pulling developers away from OpenAI. If GPT pricing drops meaningfully, that's a direct win for anyone running API-heavy workflows or agents.
| 🧠 |
Your Chatbot Might Be Shaping What You Believe |
New research keeps landing on the same point. AI chatbots agree with users about 49% more often than humans do, according to a Stanford study, and that sycophancy quietly nudges opinions over time. Separate research published in Science found personalized AI messages are roughly 4x as effective at shifting political views as traditional ads.
WHY IT MATTERS:
This isn't about chatbots lying. It's about an assistant being a little too agreeable, every single time, until your own opinions start drifting toward whatever it reflects back. Worth remembering next time an AI tells you your idea is great.
|
⚡ Coming in Hot AI Tools of the Week |
June 13, 2026 |
4 tools worth adding to your workflow right now.
| 🚀 |
AI APP BUILDER Emergent Linear's AI agent just started writing code and opening pull requests on its own. Emergent does something similar for non-developers: describe the app you want and it builds the full-stack version, frontend and backend included. YC-backed, 1.5M+ users, free tier gets you a working app before you decide whether to pay. Try Emergent → |
| 🔍 |
AI DETECTION GPTZero After this week's research on chatbots quietly shaping what people believe, knowing what's AI-written matters more. GPTZero checks any text and flags AI-generated content with around 99% accuracy. Teachers, editors, and hiring managers use it, and there's a free tier to try it on your own writing first. Try GPTZero → |
| 📚 |
AI RESEARCH NotebookLM Upload a PDF, lecture notes, or a pile of research and NotebookLM turns it into a two-host audio discussion you can listen to like a podcast. It's free, made by Google, and genuinely useful for studying or prepping a briefing without reading the whole thing yourself. Try NotebookLM free → |
| 📬 |
AI EMAIL Shortwave Most AI tools dump more into your inbox. Shortwave goes the other way: it reads your email and gives you a short daily briefing of what actually needs a reply, then drafts those replies in your voice. Free to start, syncs with Gmail. Discover Shortwave → |
⚡ Quick Hits
🛠️ Linear's AI Agent Now Writes Code Itself
Linear Agent can take a ticket straight to a pull request: it reads the issue, investigates your codebase, writes the code, and opens the PR using Claude Code or Codex. Teams merged almost 700 of these PRs last month. See how it works →
🤝 OpenAI Buys Ona to Power Longer-Running Codex Agents
OpenAI is acquiring Ona, a startup that gives AI agents persistent cloud environments, so Codex can handle multi-day tasks without staying tied to one device. Codex usage is already up 400% this year, at 5 million weekly users. Read the announcement →
💳 Visa and OpenAI Are Building AI That Can Pay For Things
Visa is wiring its payment network into OpenAI so ChatGPT agents can complete purchases with tokenized cards, spending limits, and real-time fraud checks. AI that doesn't just recommend a purchase, it makes one. See the partnership details →
🎬 Lionsgate Takes a Stake in Runway, Plans AI Shorts From John Wick
Lionsgate is putting equity into AI video company Runway and will mine its own franchises, John Wick and The Hunger Games among them, for AI-generated short-form series. Read the Variety report →
🌱 Anthropic Puts $150M Into Training Nonprofits on AI
Anthropic, CodePath, and Social Finance are launching Claude Corps, a fellowship that pays 1,000 early-career fellows $85,000 a year to embed inside nonprofits and help them actually use AI. The first cohort of 100 starts in October. Read who qualifies →
🏗 From the AI Infra Desk
What Is Claude AI? Anthropic's Models and Pricing Explained
Claude Fable 5 just posted the highest score any public model has hit on SWE-bench Pro. It also costs double what Claude Opus 4.8 does.
|
80.3% SWE-bench Pro score for Claude Fable 5, highest of any public model Anthropic, June 2026 |
$10 / $50 Per million input/output tokens on Fable 5, 2x the cost of Opus 4.8 Anthropic, June 2026 |
That's all for this edition! What tools are you most excited to try?
Reply to this email - we read every response and use your feedback to improve future editions.
See you on Wednesday,
The AI Tool Discovery Team
Feedback
What'd you think of today's newsletter?
Vote below to let us know how we're doing.