- AI Fire
- Posts
- 📱 Google Turns Phones Into AI Studios
In partnership with
Google wants app creation to feel as easy as sending a text. Its new mobile AI Studio app lets you build and launch Gemini-powered apps directly from your phone, while Meta, OpenAI, and Anthropic race toward AI agents that can remember, verify actions, and work together like software teams instead of chatbots.
What's on FIRE 🔥
IN PARTNERSHIP WITH MYOUTREACH
Misaligned sales and marketing is quietly killing B2B pipelines. Join Sandra Georgescu and Bianca Bulat for a Free Live Webinar on 27 May (11:00 BST) to get the practical playbook:
how to align on ICP
fix your pipeline handoffs
build feedback loops that actually move revenue
AI INSIGHTS
Google just announced that AI Studio is coming to phones as a standalone app. AI Studio is Google’s AI app-building tool powered by Gemini. You describe an idea, and it helps build the app for you.
Here’s what you can do:
Speak or type an app idea from your phone
Generate and preview apps instantly
Remix community projects for inspiration
Share live app links with one tap
Let builds run in the background
The iOS version launches July 1, while Android is still marked “coming soon.” Google says the app supports 76 languages and requires iOS 18+.
This feels bigger than a normal mobile app launch. Google is trying to make app creation feel as simple as sending a message. If this works well, “vibe coding” could become a mobile-first habit much faster than people expect.
PRESENTED BY SURETHING
Most AI agents demo well. Few ship real work.
Most AI agents can run a task. The problem is everything around it: setup, memory, context, cost, and figuring out what actually happened.
SureThing turns useful AI skills into autonomous agents with business context, persistent memory, cost-aware model selection, and a live dashboard. Paste a link, assign the work, and your agent reports back like a human teammate: what it did, what it cost, what needs your decision, and what happens next.
Built for founders, operators, and marketers who want AI to ship work, not become another tool to babysit.
AI SOURCES FROM AI FIRE
1. Hermes Agent Comprehensive Guide: From Zero to Your Own Open-Source AI Agent | Part 1. Just 7 weeks, it crossed 95,000 GitHub stars (now 162K stars). It supports 300+ models. Is this exactly the kind of open-source agent we need?
2. How to Give Claude Perfect Memory That Remembers Everything Permanently (Full Guide). Discover the multi-layer memory workflow that makes Claude dramatically more consistent, useful, and context-aware over time.
3. Simple Yet Powerful Claude Cowork Setup for Non-Tech Users (Full Step-by-Step System). If you don't work in tech, this surprisingly simple Claude Cowork system can replace hours of repetitive work every week.
4. Video: I Turned OpenClaw Into a REAL AI Team Workspace. Our full OpenClaw setup with multi-agent systems, secure memory, and mission control, so you can run your own autonomous AI team without wasting time or money.
TODAY IN AI
AI HIGHLIGHTS
🤖 OpenAI just gave Codex 4 major upgrades for more hands-off AI coding. You can now attach app windows, run long “goal mode” tasks for days, control locked Macs remotely, and edit websites with live annotations.
🏛️ California became the first U.S. state to formally prepare for AI job disruption. Governor Gavin Newsom signed a new order to study layoffs, worker protections, AI revenue sharing, and retraining programs.
🛰️ SpaceX’s IPO filing revealed that Anthropic is paying $1.25B per month through 2029 for access to Colossus AI supercomputers. The AI compute race is getting extremely expensive.
🛡️ Anthropic’s unreleased Claude Mythos model reportedly helped partners discover 10,000+ critical software bugs in just one month. Cloudflare alone found 2,000 bugs, including 400 high-severity issues.
🔎 Google’s new AI-first Search is already getting backlash online. Users noticed simple searches like “disregard” now return huge empty AI blocks instead of useful results, while some say Bing actually works better now.
💰 AI Fundraising & Deals: Meta may unwind its $2B acquisition of Manus after pressure from Beijing, pushing the AI startup’s founders to explore a new $1B fundraising round to buy back the company. Manus reportedly projected nearly $1B revenue this year, showing how fast the global race for agentic AI is accelerating.
HOT PAPERS OF THE WEEK
1/ AI agents can now learn reusable “skills” without retraining the model
SkillsVote from MemTensor introduces a system that collects, verifies, and evolves agent skills over time. Instead of updating the model itself, the framework improves agents by managing external skill libraries. Big result: it boosted GPT-5.2 performance on Terminal-Bench 2.0 and SWE-Bench Pro without changing the base model. This could become a new way to scale long-term AI agents faster and cheaper.
2/ Code is becoming the operating system for AI agents
Researchers from Meta, Stanford University, and University of Illinois Urbana-Champaign propose a new idea called Code as Agent Harness. Instead of code being only an output, code becomes the environment where agents think, plan, verify actions, and coordinate with other agents. Key shift: future AI systems may run more like executable software platforms than simple chatbots.
3/ AI researchers built a self-improving autonomous research system
AutoResearchClaw introduces a multi-agent research pipeline that debates ideas, fixes failed experiments, verifies results, and learns from past mistakes. The system also supports different levels of human collaboration instead of full automation. On ARC-Bench, it outperformed AI Scientist v2 by 54.7%. Important signal: the future of AI research may look more like human + AI teamwork than fully autonomous agents.
NEW EMPOWERED AI TOOLS
⚡ General Compute is an inference cloud built on ASIC chips instead of GPUs, delivering up to 5x faster AI responses for coding and voice agents.
🧪 TestSprite 3.0 uses parallel AI agents to explore and test your app like real users, with auto-heal, regression testing, and support for Claude Code and Codex.
🤖 Cleo is an AI product manager for Slack and Telegram that handles standups, follow-ups, and team decisions with transparent memory and trust controls.
📋 Nugget AI turns customer interviews into AI-generated PRDs, pain-point analysis, and dev-ready product specs connected to Linear and GitHub.
AI BREAKTHROUGH
Researchers from Meta, Stanford, and UIUC say AI is entering a new phase.
Models are no longer just chatting. They are starting to: use tools; store memory; verify actions; recover from mistakes; work with other agents
The paper calls this: “Code as Agent Harness”
This is the foundation behind tools like Claude Code, OpenAI Codex, OpenHands, and browser agents.
The key shift: Future AI won’t just generate text. It will execute tasks, remember context, and improve over time. The paper also says the real AI moat is moving away from the model itself.
The new moat becomes:
memory
execution systems
verification loops
agent coordination
That matches exactly where OpenAI, Anthropic, Google DeepMind, and Meta are heading now.
We read your emails, comments, and poll replies daily
How would you rate today’s newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello – we'd love to hear from you!
Like what you're reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team
Reply