AI in Pakistan

Walk into any market in any Pakistani city in 2026 and you will find three categories of business that together carry an enormous share of the country's economic life. The neighbourhood kiryana store, run by a family for two or three generations, sitting on a corner with the same regulars and the same suppliers. The one person freelance graphic designer, twenty four years old, working from a bedroom in Multan or Faisalabad, sending invoices to clients in Dubai and London. The mid sized exporter in Sialkot or Karachi who ships sports goods or surgical instruments or denim, runs payroll for fifty to two hundred workers, and lives or dies on the next quarter's order book. None of these three businesses look like the AI demos you see on YouTube. All three have something to gain from AI right now, today, in concrete ways, and all three have specific things AI cannot do for them and should not be asked to. This lesson walks the line carefully. The point is to leave you knowing what to actually try on Monday and what to actually leave alone.

Sketch one. The kiryana store. Owner runs it with one assistant. The owner reads slowly in Urdu, does not type fast in any language, has used WhatsApp for years, has no formal accounting system, and gets supplier price lists by SMS or by paper. What can AI do for this owner today, without disrupting his life. Three small things. First, a WhatsApp interface where he sends a photo of the price list and the model reads it, transcribes it into a clean table, and replies in Urdu with what changed since last week. Second, a simple stock alert: he tells the model what he sold today by voice in Urdu, the model keeps a running count, and on Saturday morning the model sends back a list of items that need restocking. Third, a customer message helper: when a regular asks on WhatsApp whether dal is in stock, the model can draft the reply in Urdu, the owner approves, the message goes out. None of these replace the kiryana wala's relationship with his customers. They reduce the friction of the work he is already doing. The relationship is the business. The AI is the assistant.

Sketch two. The freelance graphic designer in Multan. Twenty four years old, art college graduate, gets her clients from Fiverr, Upwork, and direct Instagram outreach. Her bottleneck is not creative ability. Her bottleneck is the unglamorous work around the design itself. Writing the proposal in clear English. Translating the same proposal into Urdu for a Karachi client who prefers the local register. Writing five client email replies a day that sound professional. Drafting an invoice. Drafting a follow up to a slow paying client. Producing fifteen variations of a logo description for the client gallery. This is exactly the kind of work AI is good at right now, in 2026, in two languages. She can move from spending four hours a day on email to spending one. Those three saved hours go into actual design, where her real value sits, and the model never produced any of the design itself. The financial outcome is straightforward. Same hourly creative output, more hours of it, higher monthly income, no loss of craft. The model worked, in this story, by handling the work that was around her work, not the work itself.

Sketch three. The Sialkot exporter. Family business, second generation, makes football and martial arts gloves for European retailers. Twenty regular foreign clients, three hundred workers, an export consignment every two weeks, an FBR sales tax filing every month, a State Bank export receipt verification process for every dollar that comes in, a labour department inspection cycle, and a permanent uncertainty about what each ministry will want next. Where can AI help. The supplier negotiation correspondence in English, yes. The first draft of a product catalog in three languages, yes. The comparison of three logistic quotes from Karachi forwarders, yes. The translation of an EU compliance document the buyer just sent, yes. Where will AI fail him, and where it would be dangerous to trust it. The exact filing requirements under the latest FBR sales tax circular, no, those change every quarter and the model will be out of date. The current State Bank export receipt verification rules, no, the rules of EFS and EERS and EPC are amended often and the cost of getting them wrong is the loss of an export rebate. The decision about whether to bring his cousin into the business as a partner, no, that is the work of his uncle, his wife, and his own night thinking, and no model is allowed in that conversation.

Pause and find a third sketch in your own life. We have given you the kiryana store, the freelancer, and the exporter. There is a fourth sketch that fits your own family or street that we cannot see from here. The kabarrhi who weighs scrap. The aunty who runs a tailoring service from her house. The cousin who runs a YouTube channel about car repair. The mother who teaches Quran from her drawing room. Pick one real person you know, and ask, where is the work around the work, what would change if that work took half the time, and where is the part that should never be given to a machine because it is the relationship or the trust or the craft. You will produce a better answer for that person than any generic AI consultant could. The point of this lesson is not that we have already worked out every business in Pakistan. The point is that you can work out yours.

What this means for the country, beyond the single shopkeeper. Pakistan has roughly five million small and medium enterprises and several million more informal traders. If even ten percent of them adopt AI for the second language, the routine ledger, and the catalog work, the productivity uplift across the country is meaningful in a way few other interventions can match, because the cost is near zero and the language barrier that used to gate this kind of help has fallen. The risk, the genuine one, is that the wrong things get automated. A model that drafts an FBR filing without a human accountant in the loop will produce a wrong filing. A model that auto replies to customers without owner approval will lose the relationship that built the business. A model that translates a legal contract for a Sialkot exporter without a lawyer reviewing it will cost more than it saved. The opportunity is large. The discipline that captures it is the same as the discipline this whole course has been training in you. AI prepares. The human decides. The cost determines who is in the loop.