The question isn't whether your AI system will hit a case it can't handle. It's where that case goes. Design the escalation path before you ship, or the first edge case becomes an incident.
Escalation is the most under-designed part of AI systems. Teams build the happy path, ship it, and discover at the first edge case that nobody defined what the agent should do when it's unsure.
The Issue Tree: When to Escalate
Branch on confidence and reversibility. High confidence, cheap to reverse: act. High confidence, expensive to reverse: act with a human checkpoint. Low confidence: escalate, regardless of cost. The matrix is the policy — write it before the agent runs.
- High confidence, reversible: the agent decides.
- High confidence, irreversible: human checkpoint.
- Low confidence: escalate, with context attached.
Who the Escalation Reaches
Power dynamics: the escalation target has to have the authority to act. Escalating to "the team" means nobody handles it. Escalating to a named role with a defined response window means the system has a real safety net.
- A named role, not a channel.
- A response window, not "when someone sees it."
- The context the human needs to decide, attached to the escalation.
What an Escalation Carries
A good escalation carries everything the human needs to decide: the input, the agent's output, the confidence, and the reason it escalated. A bad escalation carries "please review" and nothing else — which guarantees slow, bad decisions.
How FACTA Frames It
FACTA's forward-deployed and automation work builds the escalation path into the first deployment — confidence threshold, named owner, response window, and the context bundle. The system ships with a safety net, not the hope that edge cases won't happen.
Conclusion
Escalation is a design decision you make before shipping, not a feature you add after the first incident. Decide the matrix, the role, and the context — and your agent has a real safety net.
About FACTA
FACTA helps startups and growth-stage teams turn AI into production systems that keep running — not demos that impress once.
We design the architecture around the parts that actually break under real usage: tooling you own, credentials you control, failover, cost controls, observability. The boring infrastructure that keeps a system alive after launch.
Led by Matías Baglieri and Carolina Fogliato, we focus on one thing:
AI leadership that builds. Not just advises.
Ask us where your AI system escalates today.
We'll design the matrix, the role, and the context bundle it should carry. See automation governance for the ownership side.
Explore AI Automation