The Agent Harness Has a Half-Life
Every agent harness encodes assumptions about what a model cannot do. As models change, useful scaffolding can become cost, constraint, and failure surface.
Agentic Engineering
3 articles in this category.
APIs Designed for Humans Break Under Agents
Most APIs quietly rely on human judgment. Autonomous callers turn that hidden dependency into retries, side effects, and failures at machine speed. Agent-native API design makes effects, authority, recovery, and evidence executable.
Permission Is Not Control
Approval prompts ask whether an agent may take the next action. Control determines what the system can reach, how far failure can travel, and whether someone can stop and reverse it.