Exception handling
The human and system work needed for cases an automated process cannot complete normally. It is a major place where AI savings quietly reappear as cost.
Exception handling is everything that happens when a case does not fit the automated path: missing data, ambiguous requests, rule conflicts, system errors, unusual customers. Someone, or something, has to resolve it.
AI changes the exception profile rather than removing it. The routine cases get cheaper and faster, which makes the exceptions a larger share of the remaining work, and often a harder kind: they now require judgment, context and investigation. If a cheaper model produces more of them, the saving on inference may be paid back in operations.
Make it visible
Track exceptions and escalations as their own line in a cost-migration ledger with an owner, and include the effort in supervision load. Classify why each exception happens: some are inherent, others are preventable defects worth fixing at the source.
Read more in A cheaper AI model can move the cost instead of removing it.
Related terms
Escalation
Passing a case from an automated system to a person (or a higher level) because it cannot be resolved safely or well. Handled early, it is protection, not failure.
Cost-migration ledger
A small record attached to a meaningful AI change that lists, for each cost dimension, the before, after, delta and owner, so shifted costs stay visible across budgets.
Supervision load
The human attention an AI system still consumes — context supply, review, approval, correction and exception handling — measured per dependable, completed unit of work.
Used in these essays
OpenAI Decisions API turns probability into policy
The provocative part of OpenAI's Decisions API is not that AI can make choices. It is that a model score can quietly become an action — and an action can quietly become policy.
OpenAI Dots is a test of whether AI can carry a goal, not just complete a task
Dots matters less as another capable assistant than as a test of persistent delegation: can AI keep carrying a goal without giving the user a new system to manage?
The wrong questions about AI right now
Many of the questions that helped us orient ourselves around generative AI are now too blunt to be useful. The harder work is no longer asking what AI is in the abstract, but specifying where it works, where it fails, what authority it should have, and what the whole system costs.
A cheaper AI model can move the cost instead of removing it
A lower model bill can hide a higher workflow bill. The useful AI TCO question is not only what got cheaper, but where the cost moved.
The second team is the real innovation test
A successful pilot proves that one team could make something work. Organizational capability begins when a different team can reproduce the useful result without inheriting the original team's exceptional conditions.
Make the AI vendor demo fail
A polished AI demo proves that a system can succeed under prepared conditions. A buying decision needs different evidence: what happens when the system is wrong, blocked, uncertain or halfway through an action.
The first AI incident report should be incomplete
OpenAI's new misalignment disclosure framework exposes a useful enterprise design principle: record anomalous AI behavior before the organization has finished explaining it. Otherwise incident systems quietly become filters for what teams already understand.
The hidden metric in AI automation is supervision
As AI moves from assisting work to leading it, hours saved stop telling the whole story. The scarce resource shifts to human supervision: approvals, exceptions, context and judgment.