Agents run the work. You keep every decision.
Amna Labs builds AI agents for established companies. We map how work moves, build AI agents into the tools you already run, and launch them with approval controls. Every build gets one fixed price after mapping.
See how an install works, browse what we install, or read the case studies.
Six steps to a department that runs itself.
- 01
Map the department
We sit with the people who do the work and trace every enquiry, document and payment.
- 02
Set scope and price
The outcome, the boundaries and one fixed price in writing before any work starts.
- 03
Connect your stack
We build into the tools you already run. Nothing is migrated.
- 04
Run in review mode
The agent prepares real work. Your team approves every result.
- 05
Grant narrow authority
Proven actions run alone, with limits and a full audit trail.
- 06
Launch and support
Weekly reports on what ran, what closed alone, and what waits.
How a department goes live.
Every install follows the same shape. We read the operation, build into your stack, and run the result in production. No pilot that dies in a deck.
The operating audit
We read the operation before anything gets built: the repeated work, the delays, the exceptions, and what each one costs. Scope and one fixed price come out of this stage, in writing.
How the audit runs- FollowedRenewal reminder sent by hand, 3 times for one client
- FollowedSame "is my policy active" question answered 14 times
- FoundQuote compared across 3 insurer portals by copy and paste
- MappedClaims handoff with no named owner
The system, designed and priced
You see the full design before anything is built: what each agent reads, what it does alone, where it stops, and which approval sits above each action.
See the modules- ReadsPolicy schedules and CRM, read only
- Acts aloneExpiry tracking, detail requests, quote drafts
- StopsBinding cover or changing a premium
- ApprovalBroker signs off on every quote sent
Live inside your stack
It goes live in review mode, connected to the tools you already run. It earns autonomy one action class at a time, and never for moving money.
See the agents- QualifiedBuyer, 2BR, AED 2,400,000 budget, mortgage approved
- BookedViewing Thu 16:30 with the listing agent
- EscalatedPrice negotiation request sent to a person
- LoggedTranscript and fields written to HubSpot
Reported on every week
What ran, what it handled without anyone, what is waiting on a decision, and who owns it. We stay responsible for it working.
Read the case studies- Ran1,208 actions across four systems
- AloneRoutine renewals and reminders closed
- WaitingThree decisions held for the owner
- ChangedOne rule tightened after review
Pick your industry. Start there.
Departments that could run themselves.
The scope, the boundaries and the targets we put in writing, shown for three industries.
- Case 01 · Insurance
A renewal desk that never misses an expiry
A Dubai brokerage ran 1,400 motor and medical renewals a year from a spreadsheet.
Capabilities- Renewal Desk
- Quote Prep
- Policy Status Agent
- 60 days: first contact before every expiry
- Case 02 · Real Estate
Every enquiry answered in under a minute
A 35-agent brokerage received 1,100 portal and WhatsApp leads a month, most of them outside office hours.
Capabilities- Lead Qualifier
- Viewing Scheduler
- Listing Answers
- 60 sec: first reply, 24 hours a day
- Case 03 · Shipping and Trade
Document checks before customs, not at customs
A trading house moving 180 containers a month through Jebel Ali checked invoices and bills of lading by eye.
Capabilities- Trade Compliance Checker
- Exception Desk
- Market Briefing Agent
- 100%: shipments checked before filing
Case study figures are illustrative targets, not client results.
Keep your tools. We install the intelligence between them.
- Read and write, both ways.
- Every connection tested before anything depends on it.
- No forced migration, ever.
Questions before the first call.
What does it cost?
We do not publish package prices. After we map your process you get one fixed build price in writing before work starts. It changes only if the scope changes. Managed operation is scoped and priced separately, only when you want it.
How long does it take?
A single agent for one process usually runs in review mode within two to four weeks of mapping. A wider install is phased one department at a time, with the timeline in writing alongside the price.
What about our data and security?
We design for least access from day one and document every connection. Sensitive records stay in your environment, and high-risk actions always sit behind human approval.
Do we end up dependent on Amna Labs?
No. You keep the documentation, the configuration and the operating logic. Choose managed operation, or take a documented handover and run it yourselves.
What happens when an agent gets something wrong?
Consequential actions wait for a person by default, so the common failure is work waiting, not work sent. Every action leaves a record of what ran, why, and who approved it.
Do the agents speak Arabic?
Yes. Voice and WhatsApp agents handle Arabic, English, and the mixed messages customers in the Gulf actually send.
When should we not hire you?
When nothing repeats yet, when your real problem is demand, when nobody will own the agent, or when a cheap tool already solves it. We will tell you which one.