Fetching from the wire…
01
02
03
04
05
06
07
08
Agents2026-09-13 · source-backed
ModelRift ran CadQuery 2.8.0 against OpenSCAD 2026.06.12 for unattended agent-driven part generation: six agents, three tasks, Claude Opus 5 through Claude Code, one isolated subagent per cell, every output STL checked by an independent parser. All six parts printed, so the result turns on the path there. 9 of 16 failures were silent. The silent-failure count is the transferable number because it measures how much an agent can be trusted to notice its own errors in a toolchain with no verification loop, and the answer here is 44%.
Each link below shares sources, entities, or timing with this story.
This MIT library (+2,262 stars) gives Claude Code and Codex a hardware pipeline: CAD modeling with STEP output and STL/3MF/GLB export, a browser CAD/G-code/robot viewer, step.parts for sourcing off-the-shelf screws and bearings, DXF 2D drawings, URDF/SRDF/SDF robot description...
This one annoyed me, because I've been running the losing pattern. SWE-QA (arXiv 2608.01507) compares the sub-agent grep pattern that Claude Code, Codex and Antigravity all ship by default against a pre-built semantic index over the same repository. Semantic search answered 65...
Alibaba released Qwen3.6-27B on April 22. Dense architecture. Open weights. 77.2% on SWE-bench Verified, within 3.7 points of Claude Opus 4.6. On SkillsBench, it scores 48.2% versus its own 397B MoE predecessor's 30.0%. That's a 77% improvement with 14.8x fewer parameters. Let...
akitaonrails/ai-memory (Rust, 2,485 stars) hooks into agent lifecycles to capture prompts, tool calls, and session boundaries, compiling them into a git-versioned per-project markdown wiki with full-text search and optional vector retrieval. Pitch: abandon a task in Claude Cod...
holaboss-ai/holaOS puts Claude Code, Codex and its own agent in one Electron workspace where context and history live as editable local files rather than a hosted database, so memory persists across sessions *and* across agents. Built-in frontier models (Kimi K3, GLM 5.2, GPT...
Four stories about things going wrong. Here's one about something working, with actual numbers attached. In an August 7 disclosure covered by TechCrunch, Airbnb said AI now writes 60% of its new code, that concept-to-launch time on key initiatives has dropped by as much as 60%...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.