The compiler was never the first reviewer
2026.266The compiler screened a human's type errors so a human could read the rest. An agent changes who reads - Go's zero values never reached the screen.
Agents
tag:
agents
all tags →2026
What I meant to say
2026.265Arcanist made intent first-class. Claude Code moved it to the prompt. The developer went from author to auditor - and nothing enforces what I meant to say.
Certainty is a boundary too
2026.17416 rounds of adversarial prompts, verified against primary sources. The architecture that makes it impenetrable at the boundary makes it uncalibrated at the center.
The mechanism is the enforcement
2026.158Three AI code-review tools sit on the same model. The one that guarantees coverage did it with a for-loop, not a better prompt.
The archive reads back
2026.111An LLM in the middle of a fetch pipeline took orders from the file it was summarizing.
The easy half
2026.105A wallet lets you spend. That's the easy half.
The personal agent trap
2026.059A week in the agent ecosystem. Routing layers with an LLM attached, not development partners.