📡 Radar — 28 Aug 2026

The daily catch for 28 August — all days →

📡 28 Aug 2026

🛠️ AI Development

Video
MIT Missing Semester: how coding agents actually work
A popular 1-hour MIT lecture walking through how coding agents really operate — the read-edit-run-observe loop and its failure modes. A solid mental model rather than new research.
via @tetsuoai ↗
Article
A Few Pages of Markdown: committed AI config lowers the quality cost of coding agents
New study of 441 repos: coding agents lift velocity everywhere, but teams without committed AI-config files saw ~2x the complexity growth and 1.7x more warnings — config as a quality lever.
Post
Building Carbon: a coding-agent harness from scratch
Building a coding agent from scratch — Agent = Model + Harness + UI — with the interesting parts kept in view: memory, compaction, verification, subagents. On the harness layer that drives quality.
via @MyExistingBias ↗

Post
AI made ambiguity more expensive
A designer's month of agent work: the surfaces rewritten most weren't features but permission screens, defaults, logs and docs. Once people delegate to agents, vague copy becomes trust debt.
via @nielskaspers ↗

Post
AI changes the game for large IT transformation programs
A widely-shared argument that AI undercuts the safe 'nobody-gets-fired' choice of a big-name firm for major IT transformations, shifting value to intent, execution and verification. Opinion, not data.
via @chamath ↗