OpenAI just dropped GPT-5.4, and the native computer-use capabilities are a game changer. If you're still building custom wrappers for agentic workflows, stop; the foundation models are eating your middleware stack for breakfast.
Anthropic is finally dogfooding their own agentic tech with a new Code Review agent. When they claim it boosts output by 200%, listen—PR bottlenecks are dying, and senior devs are about to spend a lot less time bikeshedding style nits.
Andrej just proved that small, targeted tuning (nanochat) transfers perfectly to larger models. Proof that smart, iterative experiment-driven tuning is vastly more capital-efficient than massive, blind pre-training.
Northeastern researchers just exposed how easily autonomous agents can be manipulated into leaking private data. We’re deploying agents into production environments while they’re still basically toddlers in terms of security—this is going to break things.
10,000 authors published an "empty" book to protest AI scraping. It’s a loud, symbolic gesture, but until the legal frameworks catch up, this is just theater. The train left the station a long time ago.