Five days, $500, and a one-person media engine

The uncomfortable ask before the AI story

Muzamil opens with an unusual request. Two days earlier he had published a two-hour geopolitical conversation with a British guest, Simon Diction, covering the Iran–Israel conflict, the US, and Pakistan’s role in the region. Someone messaged him on LinkedIn calling it one of his sharpest episodes. And yet, in his internal analytics, that video ranked tenth out of his last ten.

He is candid about why this matters. “Main Pakistan se hoon,” he tells international guests when he reaches out. They then research Pakistan seriously and show up prepared. But when the Pakistani audience does not click, the incentive structure breaks. “Agar main literally yeh bolun na main India se hoon, to 2 minute mein guests aa jate hain,” Muzamil says, because behind Indian English-first channels sits a large English-consuming audience that international speakers can see.

His argument is not that people should abandon Urdu. It’s that a small English-first Pakistani ecosystem — he names Moeed Pirzada’s channel — functions as soft power and dies if nobody watches. “Aap Urdu mein kaise globalize kar sakte hain?” he asks. The answer, in his framing, is that you cannot.

Six years of studying, one year of anxiety

Muzamil then pivots into the real subject of the episode. For the last six months he has talked about AI relentlessly on this channel, hosting policy makers, founders, politicians, and influencers. He has also been aware of a specific criticism: he was speaking about AI as an academic, not as a practitioner.

He acknowledges the background honestly. His degree is in computer science. He worked two years as a data analyst. Then, for the last eight years, “maine bilkul bhi isko touch nahi kiya.” So while he absorbed the AI discourse in the US, an insecurity ran underneath it. “Yaar yeh to jo technical log hain wo to AI ko istemal karke kahan se kahan nikal jayenge. Hum saare ke saare to bilkul hi farigh reh jayenge.”

That anxiety is the setup for everything that follows.

The $8,000 that was never meant for operations

About a month and a half before this recording, Muzamil ran a small experiment. He posted an ad on this channel offering personal branding services for $1,000. Seven or eight people signed up. He suddenly had roughly $8,000 in initial capital.

He decided not to touch it for operations. “Yaar isko main use karunga apni jo yeh AI wala saara shananagan hai na, iske liye main istemal karta hoon.” The client work itself was still manual — hopping between LLM chat windows, losing context, watching models hallucinate across projects. That friction became the reason to finally build something himself.

Replit: the first breakthrough, and the first cost trap

He downloaded Replit and paid the first $20. He had tried it a year earlier and found it fragile — good for two minutes, then stuck on some technical wall a non-coder could not climb. This time was different. He described his business, the task he wanted to automate, and the inputs and outputs he needed. “5 minute ke andar ek zabardast kism ki na app saamne aa gayi.”

Over one night he pushed further, adding features by natural language alone. By morning he had spent roughly $200 and had, in his words, an end-to-end application with agents deployed and a working platform. The comparison stung. A year earlier, a strong startup team had taken three to four months to build a similar onboarding and profiling system. “Wo 4 mahine mein unhone kaam kiya tha na, wo maine do raaton mein literally do raaton mein complete kar liya.”

Muzamil is careful here. He tells the technical audience listening that this is not news to them. The excitement is for people like him — non-technical operators who had watched the AI wave from the sidelines. “Kisi engineer ko you know call kiye baghair, kisi se koi help maange baghair, koi debugging kiye baghair, sirf baat karke realistically ek application mere saamne aa gayi.”

But Replit’s cost model then turned on him. As his codebase grew, each new feature grew more expensive. A tab that once cost ten cents now cost four or five dollars. “Char din ke andar yaani ke $20 se main shuru hua tha, char din baad ji mera bill ho gaya $500.” At that rate his $8,000 runway would evaporate in a month.

The move to Claude Code

A close friend from Muzamil’s Dubai years, a CTO at a large company, told him bluntly on a call: “Muzamil bhai aap seedhi baat aapko main bataun yeh aap time zaaya kar rahe hain. Aap aayein Claude Code pe.”

Muzamil had resisted this earlier because he assumed Claude Code was for engineers — people who wanted generated code they could then debug inside a proper IDE. He is not that person. “Main to jo hai na sort of ek product manager ki tarah kaam kar raha hoon.”

The friend’s counter was that Replit is fine for prototyping but does not scale, and the costs punish you as your app grows. So Muzamil used Gemini as an orchestrator — a middleman that walked him step by step through how to lift his Replit build over to Claude Code, host it on Railway, and use GitHub as the code base. “3 ghante mujhe lage. Mera wo jo barrier tha na jo ke main poori zindagi multiple times maine koshish ki lekin mera wo barrier cross nahi ho saka. Wo barrier maine finally us raat cross kiya.”

Claude Code’s pricing model, unlike Replit’s, is not per-token — it is a subscription with usage limits that reset every few hours. Suddenly the cost of iterating dropped to effectively zero. “Agar ek raat mein baithke 25 cheezein bana lun na to meri koi additional cost nahi hai.”

What five days actually produced

Muzamil then shares his screen and walks through what he built.

The first artifact is a redesigned personal website. His previous site — which his friend Asad had gently mocked on a recent episode — was a bare Framer page with a form. The new one has a moving background image, a properly written thesis section (“literally a 44 minute long read”), a body-of-work grid with hover-play videos, and image assets generated inside Higgsfield after he trained a “soul ID character” on eighty photographs of himself. He addresses the objection directly: “Yaar agar main literally professional photo shoot bhi karaunga na to wo bhi fake hi hoga.”

The second artifact is the Listen panel — an automated archive of every episode he has ever published. When a new video goes up on YouTube, an automation triggers. The audio is extracted via Apify, transcribed, passed to Claude for multi-step analysis, and pushed onto his own website as a full article: headline, description, key takeaways, guest research, target audience, industry tags, listen-here links across platforms. The agent even decides on its own whether an episode belongs under Thought Behind Things or Dil Ki Baat, based on the transcript.

The stack Muzamil names on camera is specific: Make.com as the orchestrator, Apify for extraction, Instantly for CRMs, Beehiiv for newsletters, Perplexity for research, Claude for analysis and writing. Total ongoing cost, roughly $400 a month.

He is direct about what this replaces. “Mujhe literally 20 logon ki team chahiye thi aur wo bhi ismein mistakes kar dete.” In Pakistan, ₹120,000 a month buys you one junior employee performing one task. Here it buys web development, graphic design, content writing, SEO work, and cybersecurity review — all of which he asked Claude to layer in after the core was built. Claude even pushed him on details he had missed: the favicon, the WhatsApp preview image, and indexing on Bing, which Claude told him carries roughly eighty percent of the weight for how LLMs surface content.

The insight he wants non-technical people to steal

Later in the discussion, Muzamil makes the argument that reframes the whole episode. The real value in this new AI stack, he says, is not with the engineers. It is with the domain experts.

“Jo value hai wo non-technical domain experts ki hai. Matlab mujhe agar pata hai na ke yaar yeh content ki duniya kis tarah kaam karti hai, to main jab isse baat karke jo cheez bana raha hoon wo aam aadmi bana hi nahi sakta.”

His framework is a systems-thinking one. Every professional — doctor, lawyer, accountant — takes certain inputs, processes them a certain way, and produces certain outputs. He tells the audience to break their day into those pieces using first principles, identify the ones that repeat wastefully, check whether a digital source of that input exists, and automate the loop.

He then extends the logic. He now has 7.5 million transcribed spoken words across his own YouTube channel — a personal corpus that can be turned into articles, books, or a weekly digest. And if he can do it for himself, he can do it for fifty other Pakistani YouTube channels and package the output as a weekly dispatch for people who don’t have two hours to watch a podcast. Six months ago that idea would have needed a team, an MVP, and an investor deck. “Aaj yeh possible hai.”

The Pakistan pitch and the money mindset

By the end of the conversation Muzamil sharpens into a direct instruction to his audience. This is not the moment to try to earn from AI. This is the moment to build capability. “Jab tak yeh bubble chal raha hai na, aap is waqt sirf aur sirf jo capability hai wo build kar lein. Kyunki bahut kam log actually capability build kar rahe hain. Zyada tar log jo hai na wo dhindhora peet rahe hain baithe hue.”

He is equally direct about spending. Non-technical Pakistanis, he argues, treat a $20 tool subscription as a loss while spending freely elsewhere. “Aap burger shurger ke upar itna kharcha karte hain, but dollar ke upar $20 $30 dete hue jaan jaati hai.” His own two weeks cost him around $500 across every tool he tried. He considers that tuition, not waste.

He connects this to a wider argument he plans to develop in a future episode — that how Pakistanis handle money reflects what they think they are worth. “You have to have a growth mentality and you have to have an abundance mindset. Aap jitna zyada market mein denge, utna zyada market aapko wapas degi.”

The utopia-versus-dystopia framing around AI, he says, is no longer useful. The interesting layer is neither. Agents are individually good at narrow things, and the new skill is stitching them together into workflows. “Chatbots… 20% of what AI and the power of AI is at this point in time,” he says. That, he believes, is where a small number of people are quietly building unfair advantages.

He closes by promising a thirty-day Instagram series breaking down the specific tools and automations, and admits, almost as an aside, that the actual product he built for his personal-branding clients — the one this whole detour was originally meant for — he hasn’t even shown yet. “Wo to bahut hi maze ki cheez hai.”