Congrats on the new chapter, Andrej! Your work has inspired many of us building in AI and education.
I’ve been building a personalized AI learning product called AI-Shifu, focused on helping teachers easily make adaptive courses. I’ve always hoped to learn from your thinking on education and AI — and it feels like the window to do that may be getting smaller now 😊
Would love for you to take a quick look if you ever have time: https://t.co/kiD4d8Pac8
Powered by Gemini 3.5 Flash
Congrats on the new chapter, Andrej! Your work has inspired many of us building in AI and education.
I’ve been building a personalized AI learning product called AI-Shifu, focused on helping teachers easily make adaptive courses. I’ve always hoped to learn from your thinking on education and AI — and it feels like the window to do that may be getting smaller now 😊
Would love for you to take a quick look if you ever have time: https://t.co/kiD4d8Pac8
Powered by Gemini 3.5 Flash
Don’t believe the hype from @OpenAI , @AnthropicAI, etc. about “AI writes 80–95% of the code.”
AI writes 200%, 300%, even 1000% of the code.
Humans select what survives.
That’s how we now see human value—and which humans are truly valuable.
I am happy to share that we (the @GoogleAIStudio team) are now a sponsor of the @tailwindcss project! Honored to support and find ways to do more together to help the ecosystem of builders.
#IfAIExisted#IfAIExisted50YearsAgo
What if AI had existed 50 years ago—how would the computer industry have evolved?
A useful way to understand a new technology is to rewind to the industry’s starting point and imagine it was already there. This helps cut through today’s assumptions and see the essence more clearly.
1975 was the birth year of the PC. If AI as we know it today had existed back then, here’s how I think computing might have unfolded:
1. Keyboards wouldn’t go mainstream. Teaching everyone to type is worse than just using voice.
2. No one would worry about talking to computers disturbing others—offices already run on conversation.
3. The chatbot is the computer. Everything happens by talking to it and getting results directly.
4. AI image/video generation accelerates GUI adoption, but WIMP never needs to exist.
5. The mouse is quickly replaced by touchscreens—people draw circles on screens to guide AI, with low precision needs.
6. A chatbot can’t solve everything alone; it needs other agents.
7. Developers build agents for the chatbot to delegate tasks to.
8. A Lovart-like agent appears before Photoshop. Photoshop might never exist.
9. Word and PowerPoint never exist either—they’re just demands for text and image generation.
10. HTML/SVG emerge earlier as AI-generatable, precisely editable layout and graphics languages.
11. Spreadsheets are mostly done via real-time AI programming. Error-prone, but faster and accurate enough—just rerun if needed.
12. Programming languages stall at C/BASIC/Lisp. Training millions to code is less efficient than improving AI coding.
13. OS exist, but not for humans—only to make agents comfortable, so they’re much simpler.
14. Databases and file systems merge into an OS module for agent data storage.
15. Agent builders don’t need to code; they master prompting, context assembly, RAG, teaching AI tasks, and evaluation.
16. Customizing agents becomes a basic life skill, like using Office.
17. Enterprise software = tailoring agents to real business needs, with code only where reliability demands it.
18. Software engineering centers on requirements, judging AI-designed architectures, directing AI coding, testing, and review.
19. The internet exists mainly to let more agents collaborate—an Agentic Web.
20. Smartphones arrive earlier—they’re just chatbot terminals, much simpler.
21. Everyone’s online content or services exist as agents in the network.
Pure thought experiment. Curious to hear your takes.
Building an AI agent but not sure what to fix first?
This interactive demo helps you:
• starts from your current friction
• visualizes the minimum path that actually holds
• ends with a calm, doable 48-hour next step
It helps you stop guessing what to fix.
Try it here :https://t.co/uzFDEYxptn
Based on Top 5 Mistakes Writers Make Building AI Agents by @Nicolascole77@heymitch huge thanks to the authors.
#AgenticAI
Managers often fail due to miscommunication, not lack of skill.
Why managers fail differs by level:
New managers: unclear priorities
Mid-level: assume alignment
Senior leaders: vague, late communication upward
I tested a helpful 1-page diagnostic.
👉 Try it: https://t.co/TRtQnjNsNk
It gives you a quick, personalized AI fix.
#Leadership #communication #AIForWork