Treat the second team as a new deployment of a proven agent, not as a copy. Its cases, products, customers and systems differ, so test the agent on the second team's own past cases before it acts there, and set caps per capability for that team. That team's process owner signs off and names approvers.
Why is a second team not just more of the same?
The first team's certificate describes how the agent performed on the first team's work. A second team doing "the same job" usually has differences that matter: a different product range, a different customer base, other systems, another market's rules, or simply different habits about what counts as an exception. An agent that was excellent for retail collections can mishandle small-business collections, where the customer is a company director with personal guarantees.
The second team's process owner owns the extension for their scope. The original owner stays accountable for the first team. The agent operations lead coordinates, because changes for one team can affect the other.
What are the steps?
- Person, second team's process owner: describes their work and how it differs from the first team's: products, customers, systems, rules, volumes.
- Agent, depth 1: compares the two case sets, listing case types, fields and outcomes that appear in the second team's history but not the first's.
- Person, process owner: decides capabilities and caps for the second team. Some may be lower than the first team's; some may not apply at all.
- Person, agent operations lead: decides whether to configure separately. If the brief, sources or limits must differ, run a separate configuration for the second team rather than widening the first.
- Agent, depth 1: is tested on the second team's past cases per capability.
- Agent, depth 1: runs in shadow in the second team on live work for a period.
- Person, process owner: approves depths for the second team and names its approvers and deputies.
- Person, process owner: briefs the second team on what the agent does and does not do for them.
- Agent, at approved depths: starts in the second team, with a review date.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Describe differences | Nothing | none | Explaining the work |
| Compare cases | Lists case types unique to team two | 1 | Nothing |
| Capabilities and caps | Nothing | none | Setting them for team two |
| Separate configuration | Nothing | none | Deciding the set-up |
| Test | Answers team two's past cases | 1 | Choosing the cases |
| Shadow | Works alongside team two | 1 | Comparing results |
| Approve | Nothing | none | Depths and approver roster |
| Brief team | Nothing | none | Communication |
| Go live | Works within approved depths | as approved | Approving held items |
Which controls make the extension safe?
Certification on the second team's cases. Evidence: test results on that team's history.
Separate limits where needed. Customer limits, approval thresholds and data sources set for the second team. Evidence: configuration per team.
Two-sided change testing. A brief change made for the second team is replayed against the first team's cases as well, so neither team is surprised. Evidence: replay results for both.
Separate rosters. The second team's held items go to its own approvers. Evidence: the roster for each team.
Scope changes recorded. Inviting the agent into the second team's channels or giving it their systems is a change in what it can reach; how to connect an AI agent to your team messaging app covers keeping that narrow. Evidence: access approvals for each new source.
Which signs show it is going wrong?
- Refusals spike in the second team. The role scope fits the first team's work, not the second's.
- The first team's results change. A tweak for the second team altered shared behaviour.
- Approvers in the second team approve everything. They were not trained, or they assume the agent was already proven for them.
- Status updates confuse. The agent posts the first team's style of updates to a team that works differently; agree formats per team when you let an AI agent post updates in team channels.
- Nobody owns the second team's use. The first team's owner is assumed to cover both.
What should you measure?
Measure each team separately: agreement with staff decisions in shadow, held items and rejections, refusals by bound, corrections after the fact, and time per case. Compare the second team's figures with the first team's at the same stage, not with the first team's current figures, which reflect months of tuning.
Where does it usually go wrong?
The first team's certificate is reused. "It's already certified" is the most common shortcut and the most common cause of problems.
One configuration grows to fit both. Widening the brief to cover both teams makes it vaguer for each. Separate configurations are often cleaner.
The second team inherits the first team's task split. What stays with people differs between teams; redo that work rather than copying it.
Culture is ignored. One team welcomes the agent, the other was not asked. Managing mixed teams well is covered in how do you manage a team of AI agents and people?
Where Pryme fits
In Pryme, certification is per agent, per customer and on that customer's own cases, so the second team's use is certified on its own history. Depths can differ between teams, and each verdict names the bound that decided. This is the "widen" stage of deploying an AI agent: train, shadow, certify, widen.
Questions people ask
Can the second team start at the depths the first team reached?
Only if testing on its own cases supports them. Many firms start the second team one depth lower on sensitive capabilities and raise it after the first review.
Should both teams have the same owner?
Each team's process owner should own the agent's use in their team. If one person owns both processes, one owner is fine, but the decisions for each team should still be recorded separately.
Is a second team in another country a bigger step?
Yes. Different rules, languages and regulators make it closer to a new deployment. Re-check the rules for that market and certify on local cases.