A service level agreement (SLA) is a commitment, between a provider and its customer or between internal teams, that sets measurable standards a service must meet, such as response time, availability or accuracy, and what happens when it falls short. For AI agents, SLAs must cover both the agent's work and the human review it depends on.
Why does an SLA matter for agents?
An agent can answer in seconds. If its drafts then sit in a held queue for two days waiting for a person, the customer experiences two days. The agent's speed is irrelevant unless the whole chain, including people, is measured and resourced.
SLAs also make trade-offs explicit. A tight response target can tempt a team to raise an agent's depth so fewer items wait for review. Writing down both the speed target and the quality target forces the conversation about whether that is safe, rather than letting it happen quietly.
How does it work in practice?
An SLA for agent-supported work typically covers:
| Measure | Example target |
|---|---|
| Time for the agent to act or draft | Within minutes of a case arriving |
| Time for a person to work a held item | Same business day |
| Time to resolve an exception | Within two business days |
| Accuracy of drafts | Reviewer edits below an agreed level |
| Availability of the agent | During service hours, with a fallback route |
| Time to stop the agent when required | Within minutes |
The targets above are illustrative; set your own from your process and your obligations. How to set service levels for an AI agent covers choosing them, and the internal service level agreement template for AI agents gives a format.
What does an example look like?
A business bank's operations team agrees an internal SLA with its relationship managers for account maintenance requests. The agent drafts each change within ten minutes of the request. A servicing officer reviews held items within four working hours. Exceptions are resolved within two days. When held items start missing the four-hour target, the team does not raise the agent's depth; it adds a reviewer at peak times and the target recovers. The decision and the reason are recorded.
How is an SLA different from a service level objective?
An SLA is a commitment with consequences, often contractual. A service level objective (SLO) is an internal target a team aims for, usually stricter than the SLA so there is room before the commitment is breached. Internal agreements between teams sit in between: they are not contracts but they are owned and reported. For agents, most service levels start as internal agreements; setting service levels for work done by an AI agent explains how to make them stick.
Where do SLAs with suppliers come in?
If an agent depends on a model provider, a hosting platform or an IT service management platform, those suppliers' SLAs set a ceiling on yours. A provider outage stops the agent unless you have a fallback. When you connect an AI agent to an IT service management platform, the ticket SLAs there also start to apply to the agent's work. See our guide to deploying AI agents for where service levels fit in going live.
Where Pryme fits
In Pryme, a held action is drafted and waits for a named person, and every verdict is filed, so each decision on a held action is on record with the person who made it. Depth stays tied to Attest evidence rather than queue pressure: missing a target is never a route to more authority for the agent.
Questions people ask
Should an SLA ever justify giving an agent more depth?
No. Depth is earned through evidence on your own cases. If service levels are being missed because review is slow, fix the review capacity or the process, not the agent's authority.
What happens to SLAs when the agent is stopped?
They still apply. The runbook should say how work moves to people during an outage and what service level applies while it does.