When an AI agent hands work to a person, that person's job has changed
Zoho now ships AI agents that hand customers to a representative with full context. The handoff works. The role receiving it still has to be redesigned.
Customer service software can now pass a conversation from an AI agent to a person with the whole history attached, and let that person stop the agent before it acts. That solves the transfer. It does not solve the job on the other side of the transfer, and it is easy to give a new job to people who still have the old job description. Redesign the human role in the same project as the agent, or the blend will not deliver what it promised.
What shipped this week
On 16 November 2026 Zoho announced Blended Agents for Zoho Desk. In Zoho's words, the AI agent can "carry out tasks such as retrieving order details, checking stock and updating customer accounts", and when a case needs judgment, "the AI agent can hand over the conversation, including its history and context, to a representative." Representatives "can also pause AI execution to review or approve an action before it proceeds."
That is a sensible design. The easy reading is that the human-in-the-loop question is now settled: the agent does the routine work, a person takes the rest, nobody repeats themselves. The vendor built the plumbing, so the company only has to switch it on.
That reading skips the part the vendor cannot build for you.
The cases that reach a person are no longer a sample
Before the agent, a representative's queue was a mix: easy questions, routine updates, the occasional hard case. After the agent, the easy and routine work stays with the agent by design. What reaches the person is the residue: the ambiguous, the upset, the exceptional, and a stream of approvals.
Keep the old role around that new queue and three things go wrong:
- The old targets punish the right behavior. If a representative is still measured on cases closed per hour, every case they now get is slower by definition. The metric says the team got worse at the moment it started handling only the work that needs a person.
- Approvals turn into a rubber stamp. "Pause and approve" is only a control if the person knows what they are approving against. Without a written limit (which refunds, which account changes, up to what amount), the approval button becomes a delay with a signature on it.
- The person's judgment never reaches the agent. Every case a person resolves shows a rule the agent was missing or a case that should never be delegated. If nobody's role includes writing that down, the agent's share of the work stays where it was on launch day.
Rewrite the role, not just the routing
The question to ask is not "when does the agent hand over?" It is "what is this person now responsible for?" In a blended operation, the representative has three jobs, and each one needs to be written down.
Same person, same queue software. Different work, so different targets and training.
- Decide. Own the cases that need judgment, with the authority to resolve them. A person who receives the hard case but has to escalate it again adds a step and saves nothing.
- Approve. Check the agent's proposed action against a written limit, and reject or correct it. That limit belongs to the operation, not to whoever happens to be on shift.
- Teach. Record why the case needed a person, in a form someone can turn into a new rule or a new boundary for the agent. This is the part almost nobody staffs, and it is the part that makes the agent better each month.
Then change what you measure. Time to a correct resolution on handed-over cases, the share of approvals that were changed (zero changes for months suggests the limit could move, or the review is not real), and how many rules came back from the team this month.
- 01Agent works the caseinside written limits
- 02Handoff with contexthistory, actions, reason
- 03Person decideswith authority to resolve
- 04Reason recordedin the case record
- 05Rule updatedor kept human, on purpose
The handoff is where the agent learns its next rule, if someone's role includes writing it down.
Why this is an operations decision, not a tool setting
This is Sentynel's position in one line: redesign the operation first, then put the agents inside it. A blended agent is a change to who does what, so the people side of the operation has to be designed in the same pass, with the same care as the prompts and the integrations.
It also needs a record that shows where every case is. In Sentynel's Rodo case, WhatsApp, social and web enquiries were brought into one service operation with a visible path: New, In progress, Resolved. In the first measured window, 82% of 604 live-created enquiries reached Resolved, with a median of about one hour (pipeline records, not all conversations; no comparable earlier baseline). The point for a blended team is the stages: when a case changes hands, from agent to person or person to person, it stays in a named stage with an owner, instead of vanishing into someone's inbox.
What to do before you switch the handoff on
- Write the new role for the people who will receive handed-over cases: what they decide, what they approve, what they record.
- Set the approval limits in writing, per action, before the first approval request arrives.
- Replace the volume target for that team with time to correct resolution and rules returned.
- Book a weekly review of handed-over cases, with the person who owns the agent's rules in the room.
A question for your next team meeting: if your agent handled the routine half of the queue tomorrow, what would your best representative be measured on next week?