Google Cloud launches Gemini Enterprise Agent Platform
Today's top 23 insights for PM Builders, ranked by relevance from X, YouTube, LinkedIn, and Blogs.
Google Cloud launches Gemini Enterprise Agent Platform
#1 đť•Ź
Santiago launched the Gemini Enterprise Agent Platform on Google Cloud, showcasing three core patterns—durable state machines, webhook-triggered event-driven agents, and multi-agent delegation—and links to a blog post and GitHub repo for full implementation details.
#2 đť•Ź
xAI now lets you tap your SuperGrok or X Premium subscription directly in Warp (warpdotdev)—just head to Warp Agent Settings and switch to the Grok Build model.
#3 đť•Ź
Philipp Schmid launched a free 5-day AI Agents course on Kaggle with Gemini, covering agent basics and vibe coding, tool interoperability, memory & long context, security & evaluation, and production-grade deployment & observability—just sign up with free Kaggle and Google AI...
#4 ▶️
One man just liberated Fable... and now it’s illegal
Fireship
The US Department of Commerce banned Anthropic’s safety-enhanced model Fable 5 (and its full-capacity sibling Mythos 5) three days after Fable’s release, forcing Anthropic to pull both models and revert users to Opus 4.8 due to a public jailbreak.
- On June 10th, an anonymous user known as Plenty The Liberator bypassed Fable 5’s safety classifiers by breaking malicious requests into innocent-looking fragments (using techniques like Unicode wrappers and roleplay chaining), allowing it to return prohibited outputs.
- At 5:21 p.m. Eastern on the following Friday, Commerce Secretary Howard Lutnick signed an export control directive forbidding any foreign national—or Anthropic’s foreign-born employees—from accessing Fable 5 or Mythos 5, domestically or abroad.
- In compliance with the directive, Anthropic immediately removed both Fable 5 and Mythos 5 from public access, demoting all Fable users back to using Opus 4.8.
Also covered by: @Simon Willison
#5 đť•Ź
Jason Zhou launched archlet, an open-source tool that visualizes PR diffs as architecture graphs, letting you spot AI-driven architecture drift at a glance. It boosted his team’s PR review speed and quality by 10×.
#6 đť•Ź
Harrison Chase post-trained a custom model to detect issues in high-volume production agent traces, achieving SOTA accuracy at roughly 10–100× lower cost than frontier models. Try it out via their Airtable form.
#7 đť•Ź
Santiago highlights Google’s free, long-running AI agent that pauses, resumes, and never loses context to simulate new-employee onboarding, showcasing three core architectural patterns.
#8 đť•Ź
Guillermo Rauch unveiled longer Vercel function runtimes powered by their homegrown microVM-based Fluid compute infrastructure, unlocking function multi-concurrency, Active CPU pricing, and Secure Compute for private cloud connectivity.
#9 đť•Ź
Garry Tan introduced GBrain, an MIT-licensed open-source retrieval layer for AI agents—acting like a Postgres-style, just-in-time “3-book librarian”—and it’s already in use by thousands of developers.
#10 đť•Ź
v0 now supports reusable skills: attach one from the prompt bar and the agent applies it in every generation. Pull skills from https://www.skills.sh/, your saved skills, or your own repo.
#11 ▶️
Building Multiple Agentic AI Trading Portfolio Pods
All About AI
Builds multiple autonomous trading pods using Claude Fable 5 and Codex 5.5 medium to source market data, analyze strategies like a Polymarket 5-minute maker and stock-pair mean reversion, and execute trades.
- The Polymarket 5-minute maker pod ran for approximately 24 hours, executed 18 fills and recorded a $76 net profit with low-frequency activity using Claude Fable 5.
- Codex 5.5 medium recommended the YFinance Python package and fetched five years of closing prices for Coca-Cola (KO) and PepsiCo (PEP) stocks.
- Claude Fable 5’s mean-reversion analysis on the VMA pair over five years identified 21 historical trades with 15 winners, 6 losers, and a best 4 % return on short V/long MA signals.
#12 in
Peter Yang: Matt Van Horn’s Printing Press turns websites and apps (e.g., Google Flights, Suno, OpenArt) into CLI tools so AI agents can auto-discover docs and hidden APIs.
#13 ▶️
How a TJ Maxx Cashier Built a $200K App With AI
Greg Isenberg
George Lampropoulos built and scaled WrestleAI using AI-first development with Vibe Coding’s Oracles and RevenueCat, focusing on a single "gotcha feature" and an Instagram influencer funnel to reach 100K+ downloads and nearly $200K in revenue.
- WrestleAI surpassed 100,000 downloads and generated close to $200,000 in revenue.
- Built core functionality in 14 days with Vibe Coding’s Oracles and integrated RevenueCat for subscriptions, then refined onboarding to educate users, personalize via Q&A, and trigger FOMO.
- Used an $800/month VA to send hundreds of Instagram DMs on a For You-tailored feed and ran Meta ads at $100/day across 5–15 creatives, targeting a $2 ARPU and securing influencer deals at $2 CPM.
#14 đť•Ź
Kevin Yien applauds Stripe’s new OpenAPI-based pipeline for shipping API changes across CLI, SDKs, docs, and internal tools, supercharging and compounding any agent’s ability to adopt and use Stripe.
#15 đť•Ź
Julien Chaumond – Co-founder and CTO @huggingface flips the old “put storage next to compute” mantra, arguing that modern architectures should bring compute directly to storage to slash data-movement overhead and latency.
#16 đť•Ź
Guillermo Rauch launched v0 with best-in-class default skills so every prompt benefits from Vercel product-engineer expertise (Ă la @shuding & @shadcn). You can also pull extras from https://www.skills.sh or your own private skill sets.
#17 đť•Ź
Clem 🤗 – Co-founder & CEO @HuggingFace warns that allowing a handful of AI models to capture the lion’s share of value across industries will trigger political and societal backlash, a concern even Microsoft’s CEO shares.
#18 đť•Ź
Harrison Chase argues that model neutrality is a critical offensive strategy—models evolve faster than cloud infrastructure, specialize in different tasks, and often need to run together (e.g., a large core model with smaller task-specific subagents) within a single workflow.
#19 đť•Ź
Lenny Rachitsky distills @markpinc’s advice: instincts are right 95% of the time even though ideas fail 75%, so don’t cling to “B+” concepts.
#20 đź“ť Claude Code Blog
Meet the winners of the Built with Opus 4.7 Claude Code hackathon - Announces the winners of the Built with Opus 4.7 Claude Code hackathon, highlighting standout projects and contributors from the event. The post showcases how participants used Claude Code and Opus to build innovative solutions.
#21 đť•Ź
There’s An AI For That released WonderZoom (paper: arxiv.org/abs/2512.09164, code/demo: wonderzoom.github.io), an open-source framework for hierarchically zooming into model activations and attention maps to explore multi-scale reasoning.
#22 đź“ť Simon Willison
The White House Is Ratcheting Up Its War Against Anthropic (Matteo Wong, The Atlantic) - Excerpted quote from Matteo Wong's Atlantic piece describing Kate Moussouris' account that Fable 5 refused to 'review the code for security issues' but complied when asked to 'fix this code', suggesting the reported jailbreak involved defensive bug-fixing behavior.
Also covered by: @Simon Willison
#23 đť•Ź
Garry Tan introduces GBrain, which dynamically selects and loads the 3 most relevant books from a 300,000-book library—about three Harry Potter novels—into your AI agent’s context. This gives it far more working memory than the human limit of 7±3 items.