Topic
AI Agents and Automation
Systems that plan and act on our behalf, and the governance questions they raise.
What makes a system an “agent”
An AI agent uses a model to decide which steps to take and which tools to call — searching, writing to a database, sending a message — rather than following a fixed script. Engineering guidance from AI developers distinguishes predefined workflows from agents that direct their own process, and generally recommends the simplest design that works.
The more autonomy and access a system has, the more its errors can propagate before a human notices.
New risks, familiar controls
Security practitioners list prompt injection and excessive agency among the main risks of LLM applications: an agent reading untrusted content may be manipulated into taking actions its operator never intended.
Recommendation: limit permissions, log actions, require human approval for consequential steps, and test against adversarial inputs before deployment.
Business implications
- Accountability for automated actions must be assigned before deployment.
- Identity and access management extends to non-human actors.
- Customer-facing agents require clear disclosure.
Questions to ask
- Which decisions are we willing to delegate fully, and which require a human sign-off?
- How would we detect an agent acting outside its mandate?
- What is our rollback plan?
Insights
Articles on this topic are in preparation.
Speakers and thinkers
Sources
- Anthropic — “Building effective agents” (engineering guidance) — verified 2026-10-05
- OWASP Top 10 for Large Language Model Applications — verified 2026-10-05
- NIST AI Risk Management Framework — verified 2026-10-05
