Hmm are you seeing this with Opus 4.7 on xhigh effort and the latest version of Claude Code?
CODE
-
Modular AI Core Architecture Enables Plugin-Based Harness Extensions
By
–
Not at all. We extended core so other harnesses can be plugins now to allow more use cases (e.g. enterprises that need their preconfigured codex harness with security guardian, proxy rules etc). You can now also do silly things like use Ant SDK as harness to compare performance.
-
LLMs and the Future of Software Engineering Work
By
–
LLMs write better code than most human engineers. If code is handled, what's left for us?
— Dan Shipper 📧 (@danshipper) 22 avril 2026
I chatted with @kieranklaassen, GM of @CoraComputerCora and creator of compound engineering, on @every's AI & I about where humans fit now that AI can generate high-quality code, copy,… pic.twitter.com/md9nPAtaHQLLMs write better code than most human engineers. If code is handled, what's left for us? I chatted with @kieranklaassen
, GM of @CoraComputerCora and creator of compound engineering, on @every
's AI & I about where humans fit now that AI can generate high-quality code, copy, -
Token Role Markers Enable Context Attribution in LLM Chatbots
By
–
We convert LLMs into chatbots by using markers, eg User and Assistant: User: What's the capital of France?
Assistant: Paris.
User: What did I just say? the "I just said" attribution works because the tokens are cleanly labeled with role markers. But strip the markers and -
Claude Design: AI-Powered Design Tool with Opinionated Output Ontology
By
–
been using claude design and loving it it's basically claude code under the hood but with a very opinionated ontology for input and output – the input ontology is logo, colors, fonts, etc.
– the output ontology is slides, decks, business cards as html (with further ontology: -
Kimi K2.6: 1T-Parameter Open Model Achieves Agentic Coding
By
–
Kimi K2.6 is live on Replicate. @Kimi_Moonshot 's new 1T-parameter open model ran autonomously for 13 hours to refactor an 8-year-old trading engine — 4,000+ lines, 1,000+ tool calls, 185% throughput gain. Open weights. Agentic coding that actually finishes.
-
Build Workspace Agents with Cloud-Hosted Codex for Teams
By
–
Build workspace agents for your team, on top of a cloud-hosted Codex harness. Hook them up to tools, give them recurring tasks, and talk to them from surfaces like Slack.
— Greg Brockman (@gdb) 22 avril 2026
Easier than ever to bring the power of agents to your computer work. https://t.co/Yzf7gfHmwOBuild workspace agents for your team, on top of a cloud-hosted Codex harness. Hook them up to tools, give them recurring tasks, and talk to them from surfaces like Slack. Easier than ever to bring the power of agents to your computer work.
-
Pro image generation fails, falls back to code-generated PNG
By
–
Image generation in Pro isn’t working for me right now actually — the reasoning trace notes it has no imagegen ability, so for unambiguous image requests it spins for a while and eventually settles for making a downloadable PNG with code. Is this just me hitting usage quotas?
-
Replit Auto-Detects Issues and Proposes Fixes for Your App
By
–
When a new issue comes up, Replit notifies you as it starts preparing a fix. Review the proposed changes, hit apply, republish your app, and you’re done.