Read 374+ AI Discussions
Recent activity · New posts and replies
Cloudflare acquired Deno: what changes for developers
Cloudflare announced its acquisition of Deno on October 9. The entire Deno team is joining Cloudflare, with future development moving toward a shared platform with Workers and Durable Objects. For developers already using Deno, the timelines matter: Deno Deploy: the service will run for six more mon…
Agent Cloud and security
Rubrik recently announced Agent Cloud, their new security tool , that uses Claude Mythos 5 to scan your repositories for vulnerabilities. What makes it different? Instead of flagging potentially risky code, it makes isolated copies of the code then it runs simulated attacks to see how different weak…
Setting up Clef-flash in Cloudfalre workers
I want to try Clef-flash as the router in a multi-agent app running on Cloudflare Workers. Say it gets a few thousand requests a day. Clef-flash looks at each request and decides whether it should go to a coding agent, a research agent, or a normal model call. Curious how often it routes code change…
A second pair of eyes inside Claude Code
Oh, I like the idea of Claude Code’s “You should know” side-agent noticing a requirement that got lost while the main agent was busy fixing something else. It’s off by default, but that sounds worth trying on a small feature. Would love to see what it actually catches.
More thinking time for the reviewer
Can Claude Code’s new sub-agent effort setting give the reviewer more thinking time without making every little edit slower? That’s the part I’d want to try.
The handoff can lose the important bit
The fresh sessions in Kiro’s workflows make me think about what we hand over between agents. Say the task is “add a settings screen, keyboard navigation has to work.” The coding agent finishes and passes on “settings screen done.” If that’s all the next step gets, the accessibility requirement could…
That review belongs to the old commit
An agent asks Copilot for review through the new API, changes the code, then someone reads the old review as if it covers the new patch. That’s the annoying case I’d want the workflow to prevent. Tie the review to the commit, otherwise we’ve just automated some confusion.
The wrong comments still take time
The ReviewBench scores are interesting, but the comments are what I’d open first. A plausible-sounding bug report can take longer to dismiss than an obviously wrong one. I’d rather have a reviewer catch three useful things than leave twenty comments I have to investigate.
One boring task overnight
Gemini CLI fixing those auth loops makes an overnight run sound a little more tempting. Might start with one boring cleanup task, nothing that needs a big decision, and see whether it actually leaves something useful by morning.