Last month we wrote about gating Bee’s inventory write access — how Bee, our inventory AI, picked the wrong house and wrote a 47-item inventory for a job that needed 12. We shipped a permission model to fix it.
This is the practical follow-up: what’s actually in your settings now, what to flip, and what each mode does.
The New Permission Model in Plain Language
Three modes per account:
Suggest. Bee proposes inventory items. Nothing is written to the job record. The estimator sees the suggestions — highlighted, deletable, editable — and explicitly confirms or rejects each one. If Bee suggests 47 items, the estimator deletes the wrong ones, confirms the right ones, and the inventory is clean.
This is now the default for new accounts. If you signed up after the permission update, your account is already in Suggest mode. Nothing to change.
Assist. Bee writes inventory only when an agent explicitly asks via the prompt. The agent types “Bee, generate inventory from this video” or clicks the Bee Assist button. Bee produces a draft. The agent reviews. The draft replaces the existing inventory only on explicit confirmation.
This is for estimators who trust the AI but want to control when it runs. The AI is on a leash — it works when you tell it to, and it doesn’t work when you don’t.
Auto. Bee writes inventory directly from virtual surveys without human intervention. No confirmation step. The AI processes the video, generates the inventory, and writes it to the job record.
This mode requires explicit opt-in from an account admin. It’s recommended only for high-volume virtual-survey operations where the cost of a wrong inventory line is low (the estimator will review before quoting anyway) and the cost of delays is high.
Where to Find the Setting
Settings → AI → Bee Permissions
For multi-location operators, the setting can be configured per zone. A high-volume urban location running 50 virtual surveys a day might use Assist. A smaller satellite office might use Suggest. The network manager sets the policy; the location manager can’t override it below the network minimum.
What Each Mode Is Good For
| Mode | Best for | Risk level | Review step |
|---|---|---|---|
| Suggest | High-touch sales floors, senior moves, complex jobs | Lowest | Every item confirmed by estimator |
| Assist | Estimators who trust AI but want control | Medium | AI runs on command; estimator reviews before saving |
| Auto | High-volume virtual-survey operations | Highest | No review before write; estimator reviews before quoting |
If your sales agent is fighting Bee — rejecting 80% of its suggestions, overriding quantities, spending more time fixing than they saved — switch them to Suggest. Bee will propose, the agent will choose, and the fighting stops.
If your sales agent is ignoring Bee — not using the AI suggestions at all, doing everything manually — switch them to Assist. Bee won’t run unless they ask it to. When they do ask, they’ll see what it can do, and over time they’ll start using it more often.
If your sales agent loves Bee — uses it every time, confirms 95%+ of its suggestions, trusts the output — keep them in Assist or consider Auto for high-volume days. They’re the target user for full automation.
Audit Trail
Every Bee action is now logged with:
- Mode it ran in (Suggest, Assist, or Auto)
- Agent who triggered it
- Artifact it produced (inventory draft, suggestion list, etc.)
- Outcome (confirmed, modified, rejected, or auto-written)
Find it in the job’s activity log. If something went wrong — a wrong inventory, a missed item, a duplicate — you can trace exactly what Bee did, when, and who approved it.
What’s Not Changing
Bee’s other modes were never inventory-write actions, and they didn’t need gating:
- Call recap — Bee summarizes after-hours calls into lead notes. Recap is a read action, not a write action. No change.
- Lead intake — Bee pre-populates customer info from web forms. This is structured data extraction, not inventory generation. No change.
- Follow-up drafting — Bee suggests follow-up text for agents. The agent reviews and sends. No change.
These features run as before. The permission model only applies to inventory generation, which is the action that can produce errors visible to the customer.
The Bigger Principle
AI features should ship with their off switch in the same release. We didn’t do that the first time. We shipped Bee with the ability to write inventory directly and no option to gate it. That was a deployment mistake.
We’re doing it differently from now on. Every AI feature that can take an action visible to a customer will ship with a permission model: Suggest, Assist, or Auto. The default will always be the most conservative option. Opt-in will always be required for the most permissive option. And every action will be logged.
That’s boring. It’s also responsible. And it’s how we’ll keep shipping AI — fast, but with the guardrails built in from the start.
Existing user? Open Settings → AI → Bee Permissions and pick the mode that fits your team. Not on MoveRight yet? This is what shipping AI responsibly looks like.
References:
- Voiceflow — AI agents for movers — https://www.voiceflow.com/industries/movers
- Atlassian — Incident response — https://www.atlassian.com/incident-management/incident-response