- AI Fire
- Posts
- 🧠 AI Touches 68% of Jobs
In partnership with
Google analyzed 15 million Gemini chats and found that AI already appears across 68% of occupations, yet it supports only 21% of the tasks inside a typical job. The real impact of AI may look very different from the replacement story many people expect.
What's on FIRE 🔥
IN PARTNERSHIP WITH HUBSPOT
Want to get the most out of ChatGPT?
ChatGPT is a superpower if you know how to use it correctly.
Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.
Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.
AI INSIGHTS
Black Forest Labs just unveiled FLUX 3, a new multimodal foundation model. It was built from the ground up to generate images, video, audio, and even predict physical robot actions, all from one single neural network. Its main features:
Generates videos with native audio up to 20 seconds
Supports text to video, image to video, video to video, keyframes, and video continuation
Creates multilingual dialogue and synchronized sound effects
Maintains characters and visual references across multiple clips
Handles typography, animation, cinematic footage, and different aspect ratios
Generates and edits images with stronger prompt following and multilingual text
Can predict physical actions for robotics and other physical AI systems
In Black Forest Labs’ preliminary tests, FLUX 3 Video was preferred over Runway Gen-4.5 in 77% of comparisons, Kling v3 Pro in 60%, and Luma Ray 3.2 in 93%. These results are early and may change before the full release.
Right now, the video, audio, and action-prediction capabilities are rolling out in early access, with image generation to follow shortly.
PRESENTED BY HUBSPOT
Win AI Search Without a Big Team
92% of VCs use AI to find companies. 58% of buyers start there too. If you're not showing up in AI answers, you're invisible before the conversation even starts. Join HubSpot for Startups, Anthropic, and Marketing Against the Grain on July 16 (11am ET) for a live AEO teardown. Real startup. Real recs. Register and unlock the free Startup Visibility Bundle.
AI SOURCES FROM AI FIRE
1. Give Me 5 Minutes, I’ll Make You DEEPLY Understand & Master Kimi K3. You’ll seriously understand how it works, its architecture, and its unique capabilities. Then walk away knowing exactly how to use Kimi K3 effectively.
2. I Completely Automated AI Video Editing with Claude Code. Here's the Full Stack. A step-by-step Claude Code workflow to cut dead space, add captions, generate B-roll, create motion graphics, and render a finished MP4 with less manual editing.
3. Hermes Agent Just Got 10X Better to Fix Your Biggest Workflow Problems (5 NEW Features). Hermes Agent 0.18 unlocked 5 new features that gives AI agents memory, parallel execution, and verification, so your agentic workflows need less babysitting.
TODAY IN AI
AI HIGHLIGHTS
🎙️ Andrej Karpathy says one of his best prompting tricks is talking to AI for 10 minutes before writing anything. A messy voice ramble gives the model more context and often turns tangled thoughts into a much cleaner plan.
📈 Grok 4.5 is surging on OpenRouter, reaching 105B tokens, up 42%, and entering the top 10 ahead of GPT-5.6 Sol and Fable 5. Cost-conscious users appear to be moving over fast.
🧠 Claude Managed Agents just added effort controls, session seeding, webhooks, and support for 500 skills per session. You can now make simple tasks faster and cheaper while building much larger agent workflows.
🎤 Claude voice mode can now run on Opus or Sonnet and access Gmail, Google Calendar, Slack, Canva, and Notion while you talk. You can also switch models or move between voice and text without losing context.
✅ As AI makes fake profiles and messages easier to create, Facebook introduced Facebook Verified, a free badge confirming there’s a real person behind a profile using a short video selfie. It’s rolling out in select markets with no subscription fee.
💰 Big AI Fundraising: Etched raised $300 million at a $10.3 billion valuation, doubling in 7 months. The AI chip startup has already secured $1 billion in orders for its faster, lower-cost inference systems.
HOT PAPERS OF THE WEEK
1/ Xiaomi trains robots on over 100K hours of real-world data
Xiaomi-Robotics-1 from Xiaomi Robotics introduces a vision-language-action model trained on more than 100K hours of real robot trajectories. It can follow language instructions, handle unseen environments, and learn new tasks with limited fine-tuning data. Key result: it reaches 57.6% success on RoboCasa365, beating the previous best of 46.6%.
2/ Alibaba runs an interactive world model on one desktop GPU
ABot-World-0 from AMAP CV Lab and Alibaba Group turns keyboard actions into a continuously generated 3D-like video world. It streams 720p video at up to 16 FPS on a single NVIDIA RTX 5090, while keeping characters and scenes stable over long sessions. Big shift: interactive worlds like Google DeepMind Genie may become practical on local consumer hardware.
3/ Microsoft teaches software agents from videos and tutorials
Resource2Skill from Microsoft Research, UC Santa Cruz, and Shanghai Jiao Tong University turns tutorial videos, code repositories, articles, and examples into executable agent skills. These skills are stored in a multimodal Skill Wiki for tasks like PowerPoint, Excel, Blender, CAD, UE5, and web design. Key result: agents improved by 11.9 percentage points on average across seven software domains.
NEW EMPOWERED AI TOOLS
📱 Pushary lets you approve requests from Claude Code, Codex, Cursor, Gemini CLI, and other AI agents directly from your phone’s lock screen.
🧠 Fluree AI gives apps and AI agents trusted company context with cited answers, live structured data, and permission checks on every request.
🔥 Firecrawl /search returns the most relevant web excerpts for AI agents, delivering stronger search results while using up to 10x fewer tokens.
🚀 YC has it matches your problem with relevant companies from 4,000+ active Y Combinator startups, including pricing and integrations.
AI BREAKTHROUGH
Google’s new AI & Economy ATLAS study analyzes 15 million de-identified Gemini interactions across 150+ countries, 140 languages, 800 occupations, and 4,000 tasks. This is the most comprehensive look yet at how people are actually integrating AI into their daily routines:
AI appears in 68% of occupations, covering 90% of U.S. employment.
In a typical job, AI supports only about 21% of tasks.
Less than 10% of workplace interactions fully automate a task.
Most people use AI for research, learning, strategy, troubleshooting, and idea generation.
Manual workers also use Gemini for diagnostics, equipment checks, and technical learning.
More than 86% of interactions happen outside work, including shopping research, household tasks, taxes, licenses, and government services.
Global AI usage generally follows GDP per capita, though some middle-income countries show unusually high adoption.
English represents only around one-third of conversations.
AI adoption is already widespread, but most people currently use it as a selective assistant rather than a full replacement for human work. The smartest move right now isn't worrying about displacement, but figuring out exactly which 21% of your daily workflow can be smoothed out by a digital assistant.
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