Radar

Week 2 · Oct 12 to 18, 2026

Specs, cheaper models and safer agents

Spec-driven development keeps evolving. OpenSpec is still moving quickly, including its newer artifact-guided workflow. We’ve got an OpenSpec Deep Dive and a full Product Engineer role covering specification workflows to help you get started, plus a quick look at How Spec Driven Development Frameworks Work. We’re also showing you how to Turn a Prompt Into a Spec.

Elsewhere, cheaper models are getting seriously capable. Haiku 5.5 raises an interesting question: when do you actually need the biggest model? We explore that in When Should You Use a Cheaper AI Model?. And in Agent Memory vs Context Files, we look at whether agents actually need persistent memory, or whether giving them the right context is enough.

And as coding agents get more autonomy, what we let them access matters more too. Making Coding Agents Safe to Run covers the different ways we can protect ourselves, while How Much Access Should Your Agent Have? goes deeper into permissions, isolation and giving agents enough freedom to work without handing them the keys to everything.

Alex Garrett-SmithOct 14

This week's Versus

Should you let your coding agent remember things for itself, or write them down in a context file?Agent MemoryNotes the agent writes for itself as it works, usually after you correct it.Context FilesFiles like AGENTS.md and CLAUDE.md that you write and commit, loaded at the start of every session.Context files reach every agent on the repo and memory stays on your machine; start with a short context file and let memory handle the rest.Read the comparison

What Radar publishes

Weekly noteEvery week
VersusEvery few weeks
TLDRsSeveral a week
This Week in AI DevComing soon
The Unlearn PodcastComing soon