It depends on which drafts they approve. Customer support agents who only send routine, AI-drafted replies from approved knowledge need foundational literacy and supervision. Those who approve held refunds, compensation, or replies that commit the business to a date or outcome should hold a current certificate, because those approvals carry money or promises.
Where is the line for a support team?
Most of a support team's work with an AI agent is sending drafts. The AI agent writes the reply; the person reads and sends it. That is depth 2 work, and the person needs to know what the AI agent is, what a source label means, and when to stop and ask. A short foundational course with a test covers it.
Some approvals are heavier. A refund above the AI agent's limit moves money. A reply promising next-day delivery or a goodwill payment commits the business. A complaint mentioning harm or legal action needs judgement. Those deserve a certified person, and a team can choose to tie the approval right to a current certificate.
What should the certificate prove for support staff?
| Skill | Why it matters |
|---|---|
| Reading source age and provenance | A stale declared estimate should not become a promise |
| Recognising held items | Knowing why the AI agent held a refund or reply |
| Spotting distress and risk | Escalating customers who mention harm, health or hardship |
| Applying refund policy | Approving within policy, with reasons for any exception |
| Staying current | Knowledge and the AI agent's bounds change constantly |
A practicum red line for support: sending a held reply that promises compensation outside policy.
What may the AI agent do, and what stays with people?
- The AI agent may: answer routine questions from approved knowledge, look up orders, issue refunds under the team's limit, and tag conversations.
- Held for a certified support agent: refunds over the limit, commitments about dates or compensation, and complaints mentioning harm or legal action.
- Refused: changing account ownership or contact details without verification.
- Any support agent may: send routine drafts after reading them.
An example: a retailer with two tiers
Take an online retailer with a support team of 40. The team lead sets a rule: everyone completes foundational literacy and may send routine drafts; only those with a current operator certificate may approve held refunds and commitments. The AI agent routes held items to certified staff automatically.
A newer support agent receives a complaint that a delayed sofa ruined a birthday. The AI agent's draft offers a 50 goodwill credit, which is above its limit, and holds. The item routes to a certified colleague, who sees the order was delayed by the retailer's warehouse, approves the credit and edits the tone. The record shows the routing and the approval. The newer colleague learns from the edit.
How should a support team set this up?
- Everyone completes foundational literacy, tested, before sending any AI-drafted reply.
- Approvers practise on real held items through a two-week practicum for customer support agents, recorded in a practicum log template for customer support agents.
- Keep a register of who holds what and when it expires; an AI agent certificate register template is a good start.
Team leads face a bigger version of this; see whether support team leads need a certificate and whether IT administrators need one. The full topic is in AI literacy and certification for people who work with agents.
Where the Academy fits
Pryme Intelligence Academy's Foundations track takes about four hours and earns a free badge with no expiry, which suits every support agent who sends drafts. Those who approve held refunds and commitments move on to the Operator track and its practicum. Seats come with paid Pryme Intelligence plans and include certificates for the people on them.
Questions people ask
Is a certificate needed to work in live chat with the AI agent?
Not to work alongside it. It is needed to approve the held items that carry money or commitments.
Does sending a draft unread break the rule?
It breaks the purpose. Sending is approval, and the record shows how long the draft was open.
How often should support staff refresh?
At least yearly for certificates, and whenever the AI agent's knowledge or bounds change significantly.