Weekly: Context Engineering, ChatGPT Agent & Kimi K2
AI News for Jul 14 - Jul 21 2025
Hello! Below are 5 AI news items, 3 tools, and 1 key insight (using the 5:3:1 rule). Premium and GSD Lab members — check your additional section behind the paywall at the end. Let’s go.
📰 5 News
1. Kimi K2: Moonshot AI's open-source 1T MoE model outperforms GPT-4 on benchmarks. This open-source release boosts current AI alignment by democratizing advanced models for human oversight and customization in workflows. In the future, it may shift human work toward supervising hyper-efficient AI agents, reducing direct human involvement in complex tasks.
2. OpenAI ChatGPT Agent: Autonomous tool for complex task execution in virtual environments. It enhances present alignment efforts by allowing humans to delegate routine tasks while maintaining control via virtual interfaces. Future work could evolve to humans focusing on strategic oversight as AI handles end-to-end operations independently.
3. Google Veo3: High-res video generation model via Gemini API. Currently, it aligns AI with creative human processes by accelerating content production under user guidance. In the future, it might transform workflows where humans ideate and AI executes, blurring lines between human creativity and automated output.
4. AI Scientist-v2: Agentic system autonomously produces peer-reviewed papers. This advances alignment now by augmenting human research with AI-generated insights for validation. Future human work may involve curating AI discoveries, shifting from creation to ethical review as AI dominates scientific production.
5. Mistral Voxtral: Open-source multimodal speech model released. It supports current alignment through open-source access, enabling humans to integrate speech AI into collaborative tools seamlessly. In the future, it could redefine interactions, with humans adapting to conversational AI partners in hybrid work environments.
🧰 3 Tools
Cora — Let Cora handle your emails, so you can regain time to focus on what matters. AI email management at its finest. Up-and-coming.
Rocket.new — new kid on the AI low/no-coding tool block. Does a better job at handling errors and zero-to-one
Supercut.ai — my replacement for Loom. Super smooth video recording, near instant.
🔮 1 Insight
Prompt engineering is now no longer the cool kid on the block — it’s “context engineering.” Context engineering is about shaping the environment and background for AI systems (with say agents), while prompt engineering focuses on crafting the specific instructions or questions given to those systems (but no longer enough to guarantee results). So as things get more “agentic”, we’ll start to see more specialization (just like we see among humans) and context will become increasingly important in shaping or producing results with AI (e.g. cultural nuance).
Premium & GSD Lab Members
A little over a week before the end of July! Over the last week:
Fabrice has been making good strides with his DreamWise app. We worked through Stripe payment issues and looked at how his stack has evolved since first starting GSD Elite. He tried out Supercut for the first time, and shared an update video here. He also shared an article on how to build and run an MCP server using Docker, and why context engineering is the missing skill for building AI agents.
Grace shared a video from the co-founder of Anthropic, on AGI predictions.
Vanessa shared how she’s “touching up” Jake Gyllenhall for Prada, but also overwhelmed. We’ll work on that 💪
Rory suggested we share weekly video updates on our progress (but not everyone has gotten into that groove or rhythm yet).
I shared an article called Taste Is The New Intelligence.
New the portal:
Videos
CSD: Live Use Case (Fab). In this video, I share how I used the Stack Map template to help Fabrice through some of his challenges under GSD Elite.
Dream Visualizer (Sand Ring). In this video, I show how I use coded up a “sand ring cymatics visualizer” after dreaming about it.