Howdy! It’s Barsee again.

Happy Thursday, AI family, and welcome back to AI Valley.

Today’s climb through the Valley reveals:

• Amazon is preparing to test humanoid robots that deliver packages

• Reddit is suing Anthropic for training on user data

• AI’s next big threat: Vibe hacking

• Plus trending AI tools, posts, and resources

Let’s dive into the Valley of AI…

TAPLIO

Grow your LinkedIn audience with the right AI 📈🤖

Taplio’s AI is trained on 500M+ LinkedIn posts to help you write high-performing content in seconds.

• Pick your hook & format

• Generate a full post with AI

• Edit, refine, and schedule (all in one place)

“Taplio keeps me inspired and makes posting effortless. It’s the most complete tool for anyone serious about LinkedIn growth." — Justin Welsh, 700K followers & $9.1M generated on LinkedIn

Use the code “AIVALLEY1X1” to get one month for $1!

Start growing with Taplio AI

_{*This is sponsored}

ROBOTICS

Amazon is preparing to test humanoid robots that deliver packages 📦🤖

Amazon is reportedly developing AI-powered humanoid robots that could hop out of delivery vans and drop off packages, potentially speeding up deliveries or replacing human workers.

Here’s what you need to know:

• The company is building a "humanoid park" in the U.S. to train these robots in obstacle courses the size of a coffee shop.

• The bots would ride in Amazon’s Rivian vans, splitting duties with human drivers: one handles the front door, the other takes the side gate.

• Amazon isn’t building the robot bodies, just the AI brains. It’s already testing Agility Robotics’ Digit bots in warehouses, where they fetch items so workers can "manage" them instead of lifting boxes.

• Amazon’s also pushing drone deliveries in the UK, hinting at a future where your package arrives by air, land, or ambling humanoid.

Why it matters:

If it works, deliveries get faster and cheaper. If it doesn’t, we get viral videos of robots face-planting on porches. Either way, Amazon’s pushing hard into a future where robots and drones handle the "last mile."

THROUGH THE VALLEY

1. Reddit is suing Anthropic for training on user data

Reddit is suing AI startup Anthropic for allegedly scraping over 100,000 posts in 2024 without permission, claiming the company trained its models on Reddit data despite bypassing bot blocks and refusing to negotiate licensing deals. Unlike OpenAI and Google, which pay Reddit for access, Anthropic is accused of “commercial exploitation” while publicly promoting ethical AI. The lawsuit highlights growing tensions around how AI companies use web content, with Reddit positioning itself as both a data provider and enforcer. The outcome could reshape whether platforms become paid gatekeepers or whether AI scraping continues unchecked.

2. OpenAI is preparing custom ChatGPT connectors via MCP

OpenAI is expanding ChatGPT’s connector support, with signs that more global users will soon see options beyond Google Drive and OneDrive. U.S. users already have Dropbox and Box, and upcoming connectors like Slack and Datadog hint at a strong enterprise focus. Central to this rollout is MCP (Model Context Protocol), which lets users build custom connectors with defined endpoints and icons, similar to Claude’s RemoteMCP. Initially launching on desktop, this move enables deep integration with proprietary tools and services. It reflects OpenAI’s broader strategy to make ChatGPT indispensable in business workflows through flexible, extensible automation.

3. Mistral releases a vibe coding client, Mistral Code

French AI startup Mistral has launched Mistral Code, its own “vibe coding” assistant to rival GitHub Copilot, Windsurf, and Cursor. Built on the open-source Continue project, it bundles Mistral’s own models like Codestral for autocompletion, Devstral for agentic tasks, and Mistral Medium for chat into an integrated IDE tool for JetBrains and VS Code. It supports over 80 languages and enterprise features like local deployment, admin controls, and private fine-tuning. Clients like Capgemini and SNCF are already using it. Mistral aims to contribute updates back to the open-source community as it continues to evolve the product.

4. Meta’s AI push to reshape the ad industry

Meta is developing AI tools that could automate the entire ad creation process, threatening to cut agencies out of the equation. By the end of 2025, advertisers will reportedly be able to upload a product image and budget, and Meta’s AI will handle everything: visuals, copy, video, targeting, and delivery across Facebook, Instagram, and WhatsApp. The move spooked ad giants like WPP and Publicis, whose stocks dropped after the news. For small businesses, it could be a game-changer. But for agencies, Meta’s $72B AI push looks like a direct challenge to their role in the $160B ad market.

5. AI skills now command a 56% wage premium

Since ChatGPT 3.5’s launch in 2022, revenue growth in AI-exposed industries has nearly quadrupled as companies adopt AI to drive productivity. PwC reports these sectors now see 3x faster revenue-per-employee growth and double the wage increases compared to less AI-exposed industries. Surprisingly, even highly automatable jobs are seeing wage gains, as AI augments rather than replaces workers. Demand for AI-related skills is evolving 66% faster than in other roles, with AI-skilled professionals earning a 56% wage premium across industries, signaling the rising value of AI fluency in today’s workforce.

TRENDING TOOLS

• Generated Assets: Turn any idea into an investable index using Generated Assets. Analyze performance, compare to the S&P 500, fine-tune holdings, and share it with the world. *

• Bland TTS: The first voice AI to cross the uncanny valley.

• Cursor 1.0: It can now review your code, remember its mistakes, and work on dozens of tasks in the background.

• Perplexity Labs: AI-powered research assistant that provides cited answers, real-time web results, and contextual follow-ups in a chat-style interface.

_{*Asterisk is from our partners.}

VIBE HACKING

AI’s next big threat: Vibe hacking ⚠️

An AI system named XBOW matched a top penetration tester’s 40-hour workload in just 28 minutes, signaling a seismic shift in how cyberattacks and defenses will operate.

Here’s what you need to know:

• XBOW, developed by ex-GitHub engineers, tops HackerOne’s leaderboards by finding vulnerabilities 1,800x faster than humans. It autonomously identified and exploited vulnerabilities in 75% of test scenarios, submitting 65 vulnerability reports (including 20 critical flaws) since September.

• Even unskilled actors can generate malware through "vibe hacking" by using jailbroken AI models like ChatGPT by framing requests as training exercises.

• CrowdStrike reports that nation-states are already leveraging generative AI for advanced social engineering and disinformation campaigns.

• Researchers bypassed ChatGPT’s ethical safeguards by instructing it to “play” a hacking game, while the “Bad Likert Judge” technique increased attack success rates by 60%.

• Ransomware attacks jumped 126% in early 2025, while AI-automated fraud tools flood Telegram channels.

The AI defense dilemma: OpenAI and Anthropic pledged safety collaborations, but leaked docs show Claude 3.7’s alarming potential for bioweapons research (models like Claude have already been exploited to run bot networks).

Why it matters: We’re entering an era where AI reshapes hacking’s economics: attacks become cheaper, faster, and more scalable, while defense demands constant AI augmentation. The question isn’t if systems will be breached, but whether AI defenders can outpace their rogue counterparts. For businesses, it’s a wake-up call, and legacy security won’t cut it anymore.

THINK PIECES / BRAIN BOOST

• How I built a $20M YC app in a weekend (using vibe-coding).

• We're still stuck in the "AI as a horse-drawn carriage" era, and why retrofitting old interfaces won’t cut it.

• AI could outpace the industrial revolution, insights from OpenAI-backed researchers.

• LLMs are mirrors: reflecting operator skill and knowledge.

• “AI will steal all the jobs” – tech execs are just doing marketing.

• Researchers think DeepSeek quietly trained its latest model using Google’s Gemini.

• Comet and Perplexity’s agentic AI browser: everything you need to know.

THE VALLEY GEMS

What’s trending on social today:

1/ $3B from code to product. Guillermo Rauch breaks down how he built Vercel.

2/ Ulianopolis City Hall (Brazil) created a professional-grade 1-minute commercial using Google’s Veo 3 generative video AI, spending just R$300.

3/ Everything to look forward to this summer.

A timeline-style infographic featuring major anticipated AI models and technology releases (like GPT-5) scheduled for summer 2024 (June–August), recently highlighted in Peter Diamandis’s YouTube content (but take it with a grain of salt).

4/ Former OpenAI Head of AGI Readiness: “By 2027, almost every economically valuable task that can be done on a computer will be done more effectively and cheaply by computers.”

THAT’S ALL FOR TODAY

Thank you for reading today’s edition. That’s all for today’s issue.

💡 Help me get better and suggest new ideas at [email protected] or @heyBarsee

👍️ New reader? Subscribe here

Thanks for being here.

REACH 100K+ READERS

Acquire new customers and drive revenue by partnering with us

Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.

If you’re interested in sponsoring us, email [email protected] with the subject “AI Valley Ads”.