Are AI agents a feature, or do they need to become a production system?
Many startups are adding AI agents into existing workflows. But I’m wondering whether agents are not just a feature layer. To become real productivity infrastructure, agents may need a system around them: persistent context, tool access, permissions, human review, artifact management, and feedback loops. Without that, they may stay useful but limited. Should AI agents be treated as a full production system rather than just a product feature?