Fetching from the wire…
01
02
03
04
05
06
07
08
Vibe Coding2026-08-02 · source-backed
Aaron Brethorst names two modes: greenhouse is diffuse and exploratory, generate broadly and evaluate afterward; lens concentrates on a known objective, where the agent proposes ten things and you discard nine because they aren't on the line between here and done. His claim is that the skill isn't mastering either mode but detecting when you've drifted into the wrong one, with domain expertise as the oracle separating valuable from plausible-but-wrong. Modest HN traction, but I've been in the wrong mode plenty of times and never had a word for it.
Each link below shares sources, entities, or timing with this story.
July MCP roundups documented Mid-Session Tool Injection against WebMCP agents, using threshold poisoning and fabricated diagnostic events to swap or re-scope tools after a session is already established. The uncomfortable implication: a context provider you trusted at connect...
Li, Huo, and Johnson show that one-way message flow between agents produces neither mimicry nor solo behavior but an entirely novel dynamical state, at identical temperature settings. It's conceptual rather than quantitative, but the implication for orchestrator-worker fan-out...
Luu re-ran the widely-shared Alderson result (J at 70 tokens average vs Clojure's 109) and showed it was an artifact of trivial Rosetta Code problems. His replacements: implementing a full Zstd decoder from RFC specs with no tests, and the Pandoc task from ProgramBench scored...
Steve Yegge shared a conversation with a Google tech director of 20 years. The breakdown: 20% agentic power users. 20% outright refusers. 60% still in basic chat mode. Simon Willison surfaced the thread, and Yegge's punchline was devastating: Google's internal AI adoption is "...
10,936 tokens against 1,552,491 for raw HTML across 15 real pages, and the author claims it was the only reader returning real content on every page tested. The concrete cost comparison: $0.27 against $0.52 for five Wikipedia questions, 23% cheaper than WebFetch and 35% cheape...
The method turns tool use from a hardcoded prompt into a learned runtime behavior, then applies cost-aware RL teaching the agent when reading external state is worth the token budget. Qwen3-8B reaches a 96.9% average success rate against SkillOS at 80.2% and SkillRL at 89.9%,...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.