Pakistan can't build the LLM. It has to build the application layer.

The episode opens on a day that, by any reasonable measure, should have rearranged how anyone in an emerging market thinks about their career. OpenAI announced it was raising over a hundred billion dollars. Oracle cut thirty thousand people in a single notification, not because the company was bleeding money but because it wanted to redirect eight to ten billion dollars into AI infrastructure. And Anthropic, the company Muzamil calls “the most hottest company” in the market right now, had the source code of its Claude Code agent leak by internal mistake. Three companies. One day. Hundreds of billions of dollars in motion.

Muzamil frames this as the backdrop the rest of the conversation has to sit on top of. The world is shifting to a multipolar order. There is a sovereign debt crisis quietly forming under most major governments. Oil shortages are the first signal, but plastic, aluminium and helium shortages will follow. And then, in his words, “in the backdrop there is an industrial revolution coming.” The question he wants to put to his guest is not whether AI is real. It is whether Pakistan, sitting where it is, can actually do anything useful with the moment.

The guest is Dr. Naveed Iftikhar, co-founder of Atom Camp. PhD in urban affairs and public policy from the University of Delaware. Former governance specialist at the Ministry of Finance. Now training thousands of young Pakistanis and Saudis to work as what he calls AI-first talent. Muzamil introduces him as someone whose background, rural Punjab to bureaucracy to emerging tech, gives him an unusually grounded read on what the technology can actually do for a country like Pakistan, as opposed to “the grifty, oh, AI is gonna change the world” version of the same story.

A bureaucracy that surprised him by saying yes

Muzamil opens with the question that most listeners are probably already asking. Naveed has spent years inside the Ministry of Finance. He is now running AI bootcamps for civil servants. The two worlds, in Muzamil’s framing, feel light-years apart. Pakistani bureaucrats, he says, often look “like they’re living in the nineties.” So is this just bureaucratic theatre, or is something actually happening?

Naveed’s answer pushes back gently on the cynicism without dismissing it. The dominant perception, he agrees, is that the bureaucracy “doesn’t know anything and also they don’t do anything.” But there is also, he insists, a layer of people inside the system who are genuinely eager to learn. Off the back of Pakistan’s recently announced national AI policy, Atom Camp has partnered with the Ministry of IT and Telecom to upskill civil servants, and the surprise, for him, is that the Civil Services Academy, the very institution most people would write off as colonial fossil, was the first to embrace it. AI is now being introduced into common training programmes, into NIPA, into the National School of Public Policy. He is careful not to oversell it: “the scale is still not, you know, wider,” he says. But the door is open in a way he did not expect.

Sovereignty is not just GPUs, it is also your data

The conversation pivots into the harder question. Muzamil lays out the trap with characteristic patience. AI, in his framing, is genuinely a democratiser of intelligence. For a country that has historically lagged in collective intelligence, that is an opportunity. But there is a sovereignty trap: if all the hardware, infrastructure and foundational models sit in the United States or the Gulf, Pakistan ends up “subletting that intelligence from a third party for the rest of our lives.” Every dollar crisis becomes an intelligence crisis.

Naveed agrees, and then complicates the picture. Yes, the democratisation is real. But so is what he calls divergence, “divergence happening among nations, divergence happening among businesses, divergence happening among individuals.” Whoever has more access and more control will pull further ahead of whoever does not. And he insists that sovereignty has two layers, not one. The first is the economic layer Muzamil named: GPUs, cloud, hardware, the dollar cost of compute. The second is data. “Where that data is going? Who is using it? Who is monitoring it?” For a country with Pakistan’s security posture, that is a question that has barely been asked, let alone answered. The Gulf is building hyperscalers. Pakistan, by his read, is at “a very, very, you know, preliminary stage” on both compute and the semiconductor stack underneath it.

The five-layer cake, and the only layer Pakistan can win

Muzamil brings in Faizan Sheikh’s article about the five-point-six trillion rupee question, the piece that argues the old export model is dead and that Pakistan, as a country, has to become a product. Naveed says he read it and agrees. He reaches for Jensen Huang’s recent framing of the AI stack as a five-layer cake, energy, infrastructure, compute, models, applications, and walks through which layers Pakistan can realistically touch.

The honest answer is short. “We cannot be in that race of creating new LLM. We cannot be a new deep seek of the world, phenomena. We cannot create a new open AI.” He is direct about the local theatre around this: “sometimes Pakistanis will go and say the announce, I’ve created new LLM. This is all wrong actually. No one is making any LLM in Pakistan.” What Pakistan can credibly do, he argues, is fine-tune existing models, build small specialised models for narrow tasks, and, above all, win at the application layer.

But the application layer needs a particular kind of talent, and here he separates two things Muzamil and most observers tend to collapse. There is raw talent, the two hundred thousand engineers Pakistan produces every year, which is what Faizan is sceptical of. And there is entrepreneurial talent, which Naveed insists Pakistan does have, and which has been proven globally by the founders who built and exited abroad: the founder of Careem, the recent security exit by Rehan Jalil, the work Khalid is doing on “the internet for the future.” His point is sharp. “You don’t need, like, 500,000 entrepreneurs in the economy. Probably you need, you know, maybe fifty entrepreneurs who will change Pakistani economy.”

The problem is that those fifty currently build in Singapore, Dubai or the United States and keep only back offices in Pakistan. The work, then, is to make it possible to build from inside the country.

What “environment” actually means

Muzamil, characteristically, refuses to let the abstraction sit. “Infrastructure” and “environment” are vague terms, he says. What does it actually take? What is the institutional plumbing? He pre-empts the usual answers, the PayPal complaint he dismisses as “such a, you know, boring, redundant argument” because PayPal is a freelancer tool and the dollar problem runs much deeper than that.

Naveed lists three things, and his list is the most concrete part of the conversation. First: internet infrastructure. Sixty to seventy percent of Pakistan’s geography, by his estimate, has bad internet. He wants the government to look at its own budget honestly, comparing expenditure on motorways with expenditure on internet, and decide what era it is actually trying to build for.

Second: the banking system, and inside that, the way the State Bank treats foreign exchange around startups. He walks through a specific scenario. A founder builds a private limited company in Pakistan. A venture investor puts in one million dollars at a ten million valuation. The company grows to a hundred million. The investor wants to exit at twenty or thirty million. “Now the state bank will come. So many people will come. You know, there is a scam. Where are you taking money? You cannot take.” On paper, he says, everything looks fine. In practice, the exit becomes an interrogation, and “no venture fund will like to come to here.” That single friction, by his read, is the reason Pakistani founders incorporate offshore.

Third: public procurement. In the US, government agencies buy from startups. In Pakistan, contracting rules require billions of rupees in your account and five years of history before anyone in the public sector will touch you. A twenty-two-year-old with a working smart-metering product cannot sell it to a Pakistani electricity company. And once that procurement door is shut at home, the same founder also cannot use a Pakistani reference to go sell the same product into Nigeria or Saudi Arabia the year after. The procurement system, in other words, is quietly killing the export pipeline it claims to want.

Solve the Pakistani problem, then sell it to Africa

Muzamil pulls the thread forward. The old model, he says, was selling a mid-tier service to the United States. The new model is solving a problem in Pakistan, then taking the solution to countries that look like Pakistan: South America, Africa, Central Asia, the Far East. Naveed agrees, and reframes the country itself as an asset. “Pakistan is also a great laboratory for innovation actually because the scale is very big.” A public-sector hospital in Pakistan sees in one day what an American hospital sees in a month. The data, mostly uncaptured, sits there waiting for someone to build on top of it.

Skin in the game, and why he refuses to teach for free

The conversation turns to Atom Camp itself. Muzamil notes, approvingly, that Naveed has resisted the temptation to scale by lowering the bar. There is a barrier to entry. Students pay. Muzamil reads this as a quality signal, and Naveed explains why. There are two dominant models for skills training in Pakistan, and both produce thin results. The first is government-funded free training, which is genuinely free but, on average, “very low quality.” The second is the NGO model, run on donor money. Both, he argues, strip out the one variable that makes adult learners actually show up, skin in the game.

He references Nassim Taleb’s book by name and is unembarrassed about the economics. “If I don’t deliver quality they ask me to give back money.” Atom Camp offers monthly instalments. If a student does not like the first month, they walk and they do not pay for the next two. The structure forces him to keep the bar high. He notes, almost in passing, that the same bootcamp Atom Camp runs for around three hundred dollars sells in the US for six to ten thousand. The point is not the price arbitrage. The point is that even three hundred dollars, in Pakistan, is enough money for a student to take the work seriously.

Muzamil, who has run paid bootcamps himself, jumps in to back this up with a story that lands. He once charged students for a course, showed up to teach four-hour classes, and watched five of forty enrolled students actually attend. “Then, you know, at the end, they’ll be like, oh, you know, the course was really bad. Sir, how many times did you show up?” Free content, he points out, is everywhere, and completion rates on free courses are under one percent. The structure of payment is part of the pedagogy.

The domain knowledge layer, and who actually wins

The most useful career advice in the episode comes when Muzamil asks who the winners will be. Naveed is direct. The biggest loser, he says, is the person who learned only AI engineering or only software engineering with no domain knowledge. The traditional software-engineer-as-coder role is collapsing into a generalist who does the work of business analyst, UX designer, system architect, programmer and tester all at once. “If you are just a coder,” he says, “you are nowhere.”

The winner is the inverse profile. A civil engineer, an economist, an agricultural specialist, a public-health administrator, who layers AI on top of deep domain knowledge. AI gives them scale they did not previously have. And on top of that, he adds a third layer that he thinks is going to matter more, not less, as agents proliferate: ethics, integrity, judgement, the ability to catch a hallucination and decide whether to escalate it. “People are needed, those sort of people who can work with AI but they have their own conscience, ethics, communication and also the domain knowledge.”

Muzamil’s translation of this for the audience is clean. Spend the time building a higher-abstraction understanding of the world, then let AI do the grunt work underneath you. Be the thinking brain.

The lie the ecosystem keeps telling itself

By the end of the conversation, Muzamil asks the question he had been holding back. What lie does the Pakistani tech and startup ecosystem keep telling itself? Off-camera, he says, people are honest. On-camera they tend to give “the sort of rosy, sunny picture.”

Naveed picks two. The first is the talent claim. “We have lot of talent. This is what we say. I say we should convert it into saying we have a lot of raw talent actually.” Raw talent, in his framing, is unfinished material. Talent is what raw talent becomes after it has been shaped, polished and put to work, and Pakistan has been claiming the finished product while only producing the raw material.

The second lie, which he says is the deeper one, is about the universities. He singles out a handful, LUMS, IBA, GIKI, NUST and a few public-sector institutions, that are genuinely producing employable graduates. He notes that the University of Chicago, with whom Atom Camp runs programmes in Saudi Arabia, specifically recruits from LUMS. But across the wider system, by his read, “eighty percent of graduates from Pakistani universities are not employable.” That is the number, he says, that the sector refuses to say out loud.

Muzamil closes by repeating the position that has hardened over the last few episodes of the show. He has stopped waiting for the government. The question, for him, is what the private sector can do now, and how Atom Camp gets from two thousand students a year to twenty thousand and eventually two hundred thousand. The episode ends without a neat answer to that, which is honest. The two things the conversation has made clear are smaller and more useful: Pakistan’s window in AI is the application layer, and whether it gets through that window depends on whether a handful of structural frictions, internet, banking and procurement, get fixed before the rest of the world moves on.