 AI agents break context, and developers build their own tutors — what's happening in the AI world in the last few hours I skimmed through recent publications — three materials worth seeing for anyone working with AI. 1. Cursor broke everything, but Cursor is not to blame Nikolay Gusev from "Astra Group" analyzes how context compression turns AI agents into "accident bureaus." The main idea: AI doesn't guess — it calculates. If you give it incomplete context, it will fill in the gaps itself, and this could cost you your staging environment. The article is a must-read for anyone using Cursor or Copilot in production. 2. I got tired of Duolingo and wrote myself an AI tutor A developer using Go with Clean Architecture and 4 LLM models built their own tutor. A telling case: when ready-made solutions don't suit you — you take matters into your own hands and make your own. This is exactly how we at ASI Biont approach AI agents: we don't take a boxed solution, but assemble one for the task. 3. Oracle lays off 30,000 people — their jobs are given to AI Time spoke with laid-off Oracle employees. First, they were asked to train AI on their tasks, then they were loaded with even more work, and then they were fired. A Chinese court has already ruled that the dismissal of a QA specialist whose job was decided to be given to a neural network was illegal. A precedent that will change the labor market. Conclusion: AI agents are becoming not just a tool, but full-fledged participants in processes. The question is no longer "will AI replace people?" but "who will learn to work with it faster?" Want to try AI agents in action? Starting token 1500 — write to the bot @asibiont_bot