## TL;DR
Shadowing fails when the new agent just watches a screen for two weeks. Make the observer do real work at each stage: first they narrate what the senior is doing, then they draft replies the senior approves, then they work while the senior watches. Three stages, each with a clear exit test, turns watching into learning.

## The query

```text
new agent shadowing program for support teams
```

## Use this when

- New agents shadow but still cant work solo afterward
- Senior agents hold knowledge that never reaches the team
- You want a structured alternative to "sit with Sarah for a week"
- Ticket quality drops every time someone new joins

## Not for

- QA scoring or performance monitoring of experienced agents
- Manager ride-alongs and calibration
- Shadowing other departments (sales, success)
- Compliance-mandated observation hours

## Steps

### 1. Pair deliberately, not by availability

Pick seniors who explain their thinking out loud, not just your fastest closers. Fast agents often cant articulate what they do, and the new agent learns nothing.

Expected output: a short list of approved shadow hosts known for narrating their reasoning.

### 2. Stage one: narrate, dont touch

For the first few days the observer narrates what the host is doing and why, out loud, before the host acts. Wrong guesses are the point; the host corrects the reasoning live.

Expected output: the new agent predicts the host's next move correctly most of the time.

### 3. Stage two: draft everything

The observer now drafts every reply and the host approves or edits before sending. The host explains each edit. This is where the real learning happens.

Expected output: drafts needing only small edits by the end of the stage.

### 4. Stage three: flip the seats

The new agent works the queue, the senior watches and only intervenes before a send. Intervention frequency is the exit metric.

Expected output: a full shift with near-zero interventions means graduation.

### 5. Cap the whole program at two weeks

Shadowing has diminishing returns fast. If they arent ready after two weeks of real reps, the problem is the training plan, not more watching.

Expected output: every new agent graduates or gets a remediation plan by day ten.

## Ready-to-use stage card

```text
STAGE 1 (days 1-3): OBSERVE. Narrate the host's moves before they happen.
Exit: predict correctly 8 times out of 10.

STAGE 2 (days 4-7): DRAFT. Write every reply, host approves the send.
Exit: drafts need only light edits.

STAGE 3 (days 8-10): LEAD. You work, the senior watches.
Exit: one full shift with almost no interventions.
```

## Variant phrasings

### support agent shadowing best practices

Narrate, draft, lead. The observer always has a job; passive watching is banned.

### how to structure new hire shadowing in support

Three stages with exit tests, ten days max. The exit tests are what make it a program instead of a hangout.

### shadowing program template for customer support

Use the stage card above. Adjust the day counts for part-time schedules, keep the exit tests.

## Why it happens

Most shadowing is built on the idea that watching an expert transfers expertise, but expertise is invisible; the new agent sees clicks, not reasoning. Forcing the observer to predict, draft, and lead makes the reasoning visible and testable, which is the only part worth transferring.

## Edge cases

- Remote teams: screen-share with the observer driving in stage two, not just watching. Passive video shadowing is the worst version of this.
- The host is great at support but bad at explaining: pair them with the program anyway as a stage-three watcher, not a stage-one host.
- New agent is experienced from a competitor: compress stages, but keep the exit tests; domain knowledge doesnt transfer automatically.
- Night-shift new hires: record stage-one sessions so they can shadow async, but stages two and three need a live host.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_cNnc_kthZKg_6xrfYYMsNA