The short version
- A chatbot answers questions. A worker finishes tasks in your tools and hands back the output.
- The test: can you forward the result to a customer or accountant without redoing it?
- Ask any AI tool five questions before you pay: what does it touch, what evidence does it cite, what stops it acting, where's the audit log, what happens on a mistake?
- Halal-aware tools should make their limits explicit: process support and evidence, never rulings or certification.
What a chatbot is genuinely good at
Chatbots are excellent at explanation: summarising a policy, drafting a sentence, explaining a concept, translating. They are fast, cheap to try, and zero-risk because nothing leaves the conversation.
They are also where most small businesses stop — with a window full of good advice and no finished work. The gap between ‘here are some things to consider’ and ‘here is the screened supplier list with the certificates attached’ is the gap this page is about.
What a worker does differently
An AI worker takes a described outcome, works in the tools you actually use, and returns the deliverable — not a plan for one. Three things separate it from a better chatbot:
- It produces the artifact. A draft listing, a support reply queue, a categorised ledger, a briefing pack. Something with a file, a row, or a status attached.
- It shows its evidence. Every finding carries its source: the certificate PDF, the policy page, the line in the spec sheet. You can check the reasoning instead of trusting it.
- It stops at the boundary. Nothing is sent, spent, published or refunded without your explicit approval, and everything it did is in a log you can export.
The forwardability test
Before buying any AI tool, run one test: paste its best output into the format your customer, accountant or lawyer expects. If you have to rewrite it, you bought a chatbot. If you could send it with a glance, you bought a worker.
This is why ‘which model is it?’ is the wrong first question. Two tools can run the same model and produce completely different outcomes, because the work around the model — the mandate, the tools, the evidence, the approval gate — is the product.
Five questions to ask before you give anything access
- What exactly can it touch? Scoped, revocable access to the fields a task needs — not your whole store, books or customer list.
- What does it cite? Ask to see a sample output with its sources attached.
- What stops it acting? Approval gates on anything outbound. ‘It asks first’ should be a product constraint, not a setting.
- Where is the audit log? Every action, who approved it, when — exportable.
- What happens on a mistake? A named recovery path, and data you can delete.
Where a worker still needs you
Delegation is not abdication. A worker should never: issue a religious ruling, act as certification, give legal, tax or accounting advice, or move money on its own. It prepares; you decide. On halal questions specifically, the honest split is: the worker gathers and checks evidence, flags what needs a scholar or certifier, and says so — the ruling is not its job.
Checklist
- The output can be forwarded without a rewrite
- Every finding carries a source you can open
- Outbound actions are blocked until you approve
- Access is scoped and revocable
- An exportable audit log exists
- The tool states its limits (no rulings, no certification, no advice beyond scope)
If you would rather hand this over
AMAN is built as a worker, not a chat window: you describe the outcome in one line, it works inside a mandate you set, and the result arrives with evidence — nothing sent, spent or published without your yes. Four agents split the work (books, research, paperwork, reporting), and being halal-aware is the default rather than a feature you switch on.
Questions people actually ask
Is an AI worker the same as an AI agent?
Often, yes — ‘agent’ is the industry word for a system that can take steps rather than only reply. What matters is whether the steps end in a finished deliverable and whether an approval gate sits before anything outbound.
Do I still need a chatbot?
For explanations and thinking out loud, a chatbot is fine and cheap. The two coexist: use chat to understand a problem, use a worker to get it done.
Is it safe to give an AI tool access to my store or books?
Only with scoped, revocable access and an approval gate. Ask what fields it can read, whether it can write, and how you revoke access. A tool that needs your whole account to do one task is a red flag.
What should never be delegated?
Religious rulings, certification decisions, legal/tax/accounting advice, and any spend or publication without a human yes. A good worker refuses these by design.