AI Newsletter

AI Newsletter

đŸ€–AI Agents Weekly: MiniMax-M2.1, GLM-4.7, MiniMax-M2.1, LaMer Meta-RL, Google's 2025 AI Breakthroughs

MiniMax-M2.1, LLM Coding Workflows, GLM-4.7, MiniMax-M2.1, LaMer Meta-RL, Google's 2025 AI Breakthroughs

Dec 27, 2025
∙ Paid

In today’s issue:

  • Addy Osmani shares LLM coding workflow

  • GLM-4.7 launches with an agentic coding focus

  • MiniMax-M2.1 excels at multilingual coding

  • LaMer induces exploration in LLM agents

  • Google shares 8 research breakthroughs

  • Groq licenses inference tech to NVIDIA

  • Xiaomi MiMo-V2-Flash rivals GPT-5 High

  • Karpathy’s 2025 LLM year in review

  • OpenAI releases Skills catalog for Codex

  • Ivan Zhao on infinite minds for work

And all the top AI dev news, papers, and tools.



Top Stories

LLM Coding Workflow for 2026

Addy Osmani shares his detailed workflow for coding with LLMs going into 2026, treating AI as a powerful pair programmer that requires clear direction, context, and oversight rather than an autonomous coder.

  • Planning first: Start with detailed specifications before coding. Collaborate with the LLM to iteratively flesh out requirements into a comprehensive spec.md document - a “waterfall in 15 minutes” approach that ensures alignment before implementation.

  • Small chunks and context: Break projects into manageable tasks tackled sequentially to prevent AI confusion. Feed the AI all relevant information, including existing code, technical constraints, and API documentation, using tools like gitingest or repo2txt.

  • Testing and review: Never blindly trust AI output. Implement comprehensive testing and use AI-on-AI code reviews for additional validation. Treat AI-generated code like a junior developer’s work.

  • Customization and tools: Create configuration files (CLAUDE.md, GEMINI.md) with style guides and preferences. Switch between tools like Claude Code, Cursor, and Gemini CLI since each model has its own personality.

  • Human accountability: Stay accountable for all software produced through active participation in planning, implementation, testing, and review. AI amplifies existing expertise rather than replacing the need for strong fundamentals.

Blog

This post is for paid subscribers

Already a paid subscriber? Sign in
© 2025 elvis · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for great culture