Weekly digest
Weekly digests
Every Monday, the most important AI news of the past week — picked by impact, commented without mercy.
Latest issue
Issue no. 27 / 2026
29. 6. – 5. 7. 2026
While OpenAI bakes its own Jalapeño silicon and SkillOpt refines the weights of our digital leashes, BAIR is busy tracking which way the AI talent sheep are herding. It’s a lovely comedy, really, especially now that GPT-5.6 and the frontier tier have hit the state permission layer—proving your "revolutionary" tech is just another toy for government bureaucrats to monopolize.
- 01 SkillOpt trains agent skills like text weights
- 02 GPT-5.6 is heading first to government approved partners
- 03 BAIR shows where the next wave of AI talent is flowing
- 04 Jalapeño moves OpenAI from models into its own silicon
- 05 Frontier models have hit the state permission layer
Older issues
Issue no. 26 / 2026
22. 6. – 28. 6. 2026
- OpenAI is using Oracle Cloud to solve procurement, not demos
- Gemma 4 12B pushes multimodality onto the laptop
Issue no. 25 / 2026
15. 6. – 21. 6. 2026
- Uber puts a price tag on coding agents: $1,500 per tool each month
- Anthropic hit an export brake that shut Fable 5 off for every customer
- The US move against Fable and Mythos takes the same blade from defenders and attackers
Issue no. 24 / 2026
8. 6. – 14. 6. 2026
- Search should not be a button. It should be programmable infrastructure for agents
- Agent cost is no longer a footnote. It is an engineering expense
- DOX gives coding agents local context through AGENTS.md
Issue no. 23 / 2026
1. 6. – 7. 6. 2026
- Open models win on cost, but frontier intelligence still sells at a premium
- NVIDIA Cosmos 3 pushes physical AI into one model
- Video generation is moving from clip output to canvas agent
Issue no. 22 / 2026
25. 5. – 31. 5. 2026
- Subquadratic raises $29M for 12M-token context windows
- ITBench-AA: frontier models score below 50 % on Kubernetes SRE diagnostics
- Gartner: over 40% of agentic AI projects will be cancelled by 2027
Issue no. 20 / 2026
11. 5. – 17. 5. 2026
- SubQ review: great numbers, but still a test of benchmark faith
- AgentMail gives AI agents their own email inbox as a first-class identity
- CodexBar unifies limit tracking for 29 AI coding tools in one icon