Andrej Karpathy: 2025 LLM Year in Review
One year seems a decade in the LLM era. Gemini was still effectively at version 1.5, image models routinely failed at basic text rendering, and credible video generation had not yet arrived. DeepSeek R1 did not exist; o1 was only beginning to introduce test-time inference. The author highlights the significance of 2025 and the paradigm shift that altered the landscape.
https://karpathy.bearblog.dev/year-in-review-2025/
LangChain: State of Agent Engineering
LangChain has published a survey of 1300 professionals on Agent Engineering in the industry. Key highlights for me,
- Customer service and productivity products dominate the AI adoption.
- Quality of the output is the biggest barrier to entry for AI Agents.
- The open-source model has an equal market share between Gemini and Claude.
https://www.langchain.com/state-of-agent-engineering
Google: Introduction to AI Agents
Just like a Data Scientist, Agent Engineering is becoming the hottest job in 2026. If you’re looking to get started in this space, Google publishes an excellent overview of an Introduction to AI Agents.
https://drive.google.com/file/d/1C-HvqgxM7dj4G2kCQLnuMXi1fTpXRdpx/view
