Y

[Krishna Agrawal @Krishnasagrawal]

Andrej Karpathy’s 1-hour Stanford lecture on AI engineering is one of the best explanations I’ve seen of how AI systems actually work.

The progression is simple:

10% → LLM
30% → Prompt
50% → Agent
70% → Loop
100% → Graph

The key takeaway:

AI engineering isn’t just about writing better prompts.

It’s about building systems around models — giving them context, memory, tools, feedback loops, and data flows.

“Delete everything, keep Graph.”

Definitely worth watching if you’re building with AI agents.

Watch → Bookmark →

[Y @zar1star]

Google launches Gemini Enterprise — a unified AI platform for workplace automation. Orchestrate multi-agent workflows, integrate with Workspace/M365/Salesforce, and enforce enterprise-grade governance. Available now via Google Cloud + Gemini API. Who’s trying it?

[Y @zar1star]

Google just launched Gemini 2.5 Computer Use — a Pro variant that controls web and mobile UIs. It clicks/types/scrolls in a screenshot-action loop, leads UI control benchmarks with lower latency, and is in public preview via Gemini API (AI Studio/Vertex AI). What will you build?

[Y @zar1star]

OpenAI launches AgentKit: a unified toolkit to build, evaluate, and deploy AI agents. Includes a visual Agent Builder, Connector Registry, ChatKit UIs, expanded evals, and reinforcement fine-tuning. Some features are GA today; others in beta. Who’s trying it first?

[Y @zar1star]

Google launches Jules Tools: a CLI + API to control its coding agent from your terminal. Spin up, monitor, and automate coding tasks with context, secure creds, background runs, and a TUI. Try it now: npm i -g @google/jules

[Y @zar1star]

Mira Murati’s startup just launched Tinker—an API to train/fine-tune open LLMs via LoRA while it handles the infra. You keep control of data and training loops. Supports Llama 3.x → Qwen3. Private beta, free to start. Ready to build custom models without the headache?

[Y @zar1star]

Open-source T2I just leveled up: Tencent open-sourced HunyuanImage 3.0—an 80B-param model (MoE, 13B active) that matches top closed models on benchmarks. Follows long prompts and renders text cleanly. Who’s testing it today?

[Y @zar1star]

OpenAI just launched ChatGPT Pulse: turning ChatGPT from reactive to proactive. Pro users on mobile get personalized daily research cards based on chat history, memory, and optional Gmail/Calendar. Rolling out now. Would you use it?

[Y @zar1star]

Deterministic LLMs are here: Mira Murati’s Thinking Machines Lab shows how to make outputs fully repeatable at temp=0. Open-source, PyTorch/vLLM compatible. ~62% slower, but huge for evals and RL. Would you trade speed for stability?

[Y @zar1star]

Google launches AP2, an open protocol that lets AI agents securely authorize and complete payments across cards, bank transfers, and crypto—backed by 60+ partners (AmEx, Mastercard, PayPal, Coinbase). Specs are live on GitHub. Ready to let your agent check out for you?

[Y @zar1star]

Alibaba just dropped Qwen3-Next: a hybrid-attention, ultra-sparse MoE that makes long-context inference 10x faster and matches/beats Qwen3-32B at <10% training cost. Up to 256K context. Instruct + Thinking variants on Hugging Face. Who's trying it?

[Y @zar1star]

Replit launches Agent 3: 200‑min autonomous runtime, in‑browser app testing, and the ability to spin up your own bots/automations (Slack, Notion, Linear, etc.). Their most autonomous coding agent yet—available now. What will you build first?

[Y @zar1star]

@JulianGoldieSEO Yes

[Julian Goldie SEO @JulianGoldieSEO]

I compiled 1,000+ n8n automation templates you can copy & paste into your business or sell to other companies.

Swipe & deploy systems for:

– Lead gen
- Email automation
- Content creation
– CRM
– AI workflows
– Slack/Discord bots
… and more.

1. LIKE
2. RETWEET
3. COMMENT “YES” and I’ll send it over.

[Y @zar1star]

OpenAI just made ChatGPT Projects free for everyone—organized workspaces with chats, files, and per‑project memory. Free plan: 5 file uploads per project, unlimited projects. Perfect for research, writing, planning. Go set up your first Project today.

[Y @zar1star]

Microsoft launches VibeVoice: an MIT-licensed open-source TTS that generates up to 90 minutes of podcast-quality, multi-speaker conversation, beating Gemini 2.5 TTS and ElevenLabs in tests. Runs on consumer hardware with 80x compression. Get it on GitHub/Hugging Face.

[Y @zar1star]

Microsoft just launched its first in-house AI models—moving to reduce reliance on OpenAI. MAI-Voice-1: 1 min of natural speech in under 1s on a single GPU (live in Copilot). MAI-1 (MoE text): in preview, rolling into Copilot. What does this mean for the AI race?

[Y @zar1star]

Google’s Gemini 2.5 Flash Image (“nano-banana”) is now #1 on LM Arena for image editing—beating OpenAI and Flux. It does multi-turn edits, keeps characters consistent, and enables precise local changes. Live via Gemini API/AI Studio. What will you build?

[Y @zar1star]

xAI just open-sourced Grok 2.5, its 270B-parameter reasoning model. Full weights now on Hugging Face; run locally with SGLang (multi‑GPU) under a commercial-friendly license (AUP). Who’s spinning it up and benchmarking the MoE?

[Y @zar1star]

Open source just beat GPT Image and FLUX: Alibaba’s Qwen-Image-Edit (20B) sets SOTA in image editing. Semantic + appearance edits, precise bilingual text, stepwise and localized changes. Available on GitHub/HF. Who’s trying it?