Fetching from the wire…
Vibe Coding2026-09-26 · source-backed
Ayman Nadeem's September 24 post (368 points, 332 comments) says building Nuanced taught him that "planning != plan": users didn't value the spec artifact, AI-written specs are tedious to read, and demanding a finished plan before building broke how people actually interleave thinking and doing. He argues for an understand-act-inspect-clarify-adjust loop. The open problem he names is keeping humans able to understand a system as agent count goes from five to hundreds. The comment thread is split, which is the right state for this question.
Each link below shares sources, entities, or timing with this story.
Three releases in about eight hours on September 24, 1,060 stars gained today against 17,106 total. The 1.0 release adds a Server SDK and an AI SDK letting agents read and edit documents, update formulas, generate charts and visually inspect the result. It also removes APIs, s...
MIT, Python, appeared September 20, converting yes/no, pick-one and rubric-score evaluation into calibrated probabilities in a single forward pass. Claimed: a few thousandths of a cent per trace, p50 244ms and p95 371ms against several seconds for the same eight metrics throug...
LeonardLeroy/i-dont-believe-you (September 15) is "a skill to stop your coding agent from telling you it works when the diff says otherwise." coldteadotai/abide (September 18, 105 stars) exists to make a coding agent obey project rules it already has. kunchenguid/compact-advis...
Browserbase rebuilt it so the agent drives the browser from an extension loaded at startup instead of over CDP through Playwright, cutting round-trip time and CDP race conditions. Browserbase The repo has 24,475 stars with a September 18 push, but the newest tagged releases ar...
Roland Gao published GoBench on September 15, scoring frontier models against a calibrated ladder of KataGo opponents. Astra Max 2,568, Astra High 2,227, Claude Opus 5 High 2,076, GPT-5.6 Sol Max 1,929, against KataGo's 4,400. Given coding tools and two hours of preparation be...
Marcel Pociot's experimental app, published September 14 at 173 stars two days later, routes system-level Siri and Shortcuts requests to a local Claude Code session with its tools and file access instead of to Apple's own model. Small repo, but the first concrete example of Ap...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.