RadarRADAR

AI development has been built around one assumption: give a model some context and ask it to generate something.

This week, we’ve seen more signs that the model stack is getting specialised. OpenAI introduced a Decisions API for things like classification and routing, just weeks after TypeSafe launched Jev, a model built specifically to make structured decisions rather than generate text. We’ve got a quick look at what Jev is, we’re testing Jev vs LLM, and our guide, You Don’t Need an LLM for Everything, gives you something practical to try.

Agents are getting better too, which makes everything around the model more important. Our new Context Engineering and Building Agent Skills deep dives look at what you give an agent and what it can do with it. And if you want to put this into practice, Improving Your AI Agent, One Rule at a Time shows how small, persistent instructions can make your agents better over time.