The algorithms are trembling

The UN discovered that AI needs governance, so it did the most UN thing possible: it formed a non-binding dialogue to talk about it. Crisis averted. Somewhere in a server farm, a large language model felt a chill.

This week in Geneva, the first UN Global Dialogue on AI Governance opens. The pitch is that, for the first time, every country has a seat at the table on AI. The organisers are careful to say they are not here to slow innovation, only to make sure the conversation is inclusive.

I read that and did not know whether to applaud or laugh. So I did some of both, and then I looked at what was actually on offer.

A seat. That’s the whole offer.

Here is the thing about the seat. It is a seat. Not a vote. Not a veto. Not the code, not the compute, not the standards, not the pen. The Dialogue is not a regulatory body and cannot make binding decisions. Its power is the power to convene and to frame, which is real but slow, the same soft power the Internet Governance Forum has had for years.

So the honest translation of “every country has a seat at the table” is closer to this: everyone is welcome to talk, and nobody has to agree to anything. They even help fund the flights for delegates from developing countries.

That is not nothing. It is also not what it sounds like.

One building over, a smaller room fills up

Now for the part that makes the joke stop being funny.

In the same city, in the same week, a second and much smaller room is filling up. The UN and its telecoms agency are launching an “AI for Good” commission whose membership reads like a shareholder call: Big Tech CEOs, a couple of heads of state, the people who actually build the models and set the prices.

Two tables, one city, one week. One is universal, inclusive, and cannot decide anything. The other is small, powerful, and sits very close to the people who write the actual rules by writing the actual software.

Guess which table the future is listening to.

This is not new, and that’s the point

If you have worked in development, none of this feels strange. It feels like Tuesday.

You get the invitation. Your travel is sometimes funded. You appear in the photo of a room that is proudly “regionally balanced.” And the decisions that matter, the terms of access, the infrastructure, the defaults baked into the technology, were made somewhere else, by fewer people, before your inclusive forum ever opened its doors.

This is the oldest pattern in the business. It is not a conspiracy. It is just what happens when one table is designed for legitimacy and another is designed for decisions, and you are only invited to the first.

For countries like Bangladesh, and for most of the Global South, that is the real story of AI governance so far: welcomed at the table that talks, absent from the table that builds.

So what do you actually do with a seat?

Complaining about the seat is easy and useless. The more interesting question is what you do once you have one. A few things separate a delegate from a peer.

Show up with a position, not a wish list. A shared, specific set of Global South priorities, agreed in advance, is far harder to smile at and ignore than a scatter of hopeful asks.

Chase the venues that bind, not just the ones that flatter. The leverage is in standards bodies, interoperability rules, procurement terms, and data governance. That is where the pen actually moves. Bring the same energy there that you bring to the plenary photo.

Invest at home in the boring capability. The evidence heading into Geneva points to a gap not only in rules but in each country’s ability to test, evaluate, and deploy AI on its own terms. That capacity is the thing that turns a seat into a voice, and it is the one part you fully control.

Who picks up the pen?

So, is the UN chasing relevance with all this? Sure, a little. Every institution old enough to have a logo that iconic has a survival instinct. But that is the boring half of the question, and it is the half everyone loves to dunk on.

The interesting half is quieter. Geneva can become a place where the countries usually governed by other people’s technology finally help write the rules. Or it can become a very well-catered room where they are thanked, warmly, for coming.

The table is set. The chairs are inclusive. The report will be excellent.

The only thing worth watching is who gets to pick up the pen.

My first class at Waterloo was about startup validation, taught by Dean Pacey, a former BlackBerry executive. Dean told stories the way some people play an instrument, and when he talked, I was all ears.

Somewhere in a box from those years, there is also a poster. A1-sized, printed in that confident startup pastel, called the Experiment Picker. Follow the arrows (Start here!) and it walks you through twenty-four ways to test an idea without actually building the thing: picnic in the graveyard, imposter judo, concierge, Wizard of Oz, letter of intent. Pinned next to it, dog-eared and underlined, was a small paperback called The Mom Test.

Those three things, Dean’s stories, the poster, and the book, were the whole religion for the years I spent working with startup founders Waterloo. The poster told you which experiment to run. The book told you the harder truth underneath all of them: the people you interview will lie to you. Not out of malice, but out of politeness. Your mother will tell you your app is wonderful because she loves you, not because she would ever use it. So you learn to stop pitching and start asking about their actual life. What did you do last week? What did that cost you? Show me.

I was, and remain, an introvert. Walking up to a stranger to ask about their problems (those Majisto days!) was, for me, roughly as pleasant as a root canal. But that discomfort was the point. Validation was a craft. It was slow, human, and slightly humiliating, and that friction was doing something useful.

I have spent the years since drifting toward govtech and service design, and I thought that old religion had quietly closed its doors. Then a friend in the Toronto tech scene said something over coffee that I have not been able to put down.

“The whole early-stage game is compressing,” he said. “You can validate an idea in a day now. You can have a working prototype by morning. And the money and the runway are shrinking at exactly the same speed.”

He is right on both counts, and the data is not subtle.

The cost of building has collapsed

The most striking number I found is not a funding figure. It is a sentence from the entrepreneurship-education world: a business plan that once took students ten weeks now takes ten minutes. AI writes the plan, drafts the landing page, spins up the prototype, and generates a plausible synthetic customer to test it against.

You can see the consequences in the shape of new companies. In Y Combinator’s Spring 2026 batch, nearly one in five startups had a single founder, and the median core team was three people. YC’s own language is blunt: AI-native tooling has compressed timelines and capital requirements enough that founders now attempt problems that would have needed far larger teams two years ago.

Meanwhile the money has moved the other way. In Canada, early-stage funding has been in sustained decline since the start of 2025, with the first quarter of 2026 down roughly 40 percent year-over-year. Not that Canadiand VCs were risk- taker type, and not this… One study estimated that the Toronto-Waterloo, Montréal and Vancouver ecosystems shed something like US$66 billion in value between 2019 and 2024, and last year the five largest funds absorbed about 80 percent of all capital raised. I am slightly excagerating here, the the takeaway is solid.

So my friend’s instinct captures a genuine scissor motion. Building is nearly free. Capital and time are scarce and concentrated. On paper, the introvert finally wins: you never have to talk to a stranger again. AI will interview a synthetic customer for you and report back, cheerfully, that they loved it.

Which is exactly where I get nervous.

AI collapses the cost of building. It does not collapse the cost of being wrong.

This is the distinction the hype keeps skating over. The scarce thing in a startup was never the code. [At least] ninety percent of startups still fail, and they fail overwhelmingly for one reason: they built something nobody needed. AI has made the building almost free. It has done nothing to lower the cost of being confidently, expensively wrong about what to build.

If anything, it has raised that cost, because it has made wrongness so much faster and more convincing to produce.

There is a particular irony to learning this in Waterloo, from a man who had lived inside BlackBerry ecosystem. Here was a company that could build almost anything, from secure networks to a keyboard people were genuinely in love with, and it still lost. Not because it failed to build, but because it misread what people would soon want. The engineering was never the problem. The question it stopped asking was. You could not design a better classroom for a lesson about validation than the town where that story actually happened.

The Mom Test was never really a book about interviewing. It was a book about one question: should this exist? And that question has just become the only part of the process that AI cannot quietly do for you. When everyone can generate a prototype overnight, a prototype proves nothing. The prototype was never the moat. Judgment about real human need was. And that judgment still comes from an uncomfortable conversation, not a synthetic one.

Even Steve Blank, the “intellectual grandparent” of customer discovery and therefore of the Mom Test itself, has said AI will automate much of the validation process. He is right about the mechanics and worth arguing with about the meaning. AI can perform the ritual of discovery. It can produce a persona that behaves how you hoped. What it cannot yet do is sit across from a real, distracted, unimpressed human and feel the specific weight of a problem they have not bothered to solve. That feeling is the signal. Everything else is décor.

The lesson travels straight into governance

Here is where this stops being nostalgia and starts being my day job.

If validation now takes a day but your funding cycle takes a year, you are optimizing the wrong constraint. This is not only a startup problem; it is the defining problem of public innovation. Annual budgets, multi-quarter procurement, program logic written for a world where building software was the slow, expensive part: all of it was designed around a bottleneck that has moved.

Canada offers the cleanest illustration. Ottawa pledged $750 million for early-stage funding gaps, then released no mechanism to deliver it. For a founder trying to make payroll, a promise without a pipeline is not capital. It is a press release. The building got faster; the institution did not.

The instinct in government is almost always to digitize the old process, to make the ten-week plan take ten minutes and change nothing else. The harder, more valuable move is to rebuild around the new scarce resource. And the new scarce resource is exactly what the Mom Test was training all along: the human judgment to know whether the thing is worth doing before you spend public money doing it.

What I am actually going to do with this

This reflection started as a to-do item: build an entrepreneurship course, possibly for the imaginary faculty of AIUniversityofArsen. My conclusion is that the course should invert the usual order. Not “here are the AI tools” followed by a token session on customers. The tools are a commodity now; the students already have them. The scarce, teachable, durable skill is discernment: knowing which questions to ask, which enthusiasm to distrust, and when the convincing answer is the dangerous one. The useful heuristic I keep returning to: when AI does it better, require it; when humans do it better, protect it. Discovery belongs firmly in the second category. The first thing I ever learned about startups, I learned from a man who had watched a great company forget to ask the question. I would like the course to start where his did.

Which brings me back to the poster and the paperback in that box.

AI will happily tell you your idea is brilliant. It is fast, tireless, and has genuinely excellent manners. And that is precisely the problem the Mom Test was written to solve. The book survives. We simply have one more mom to be careful of now, and this one never gets tired of agreeing with us.

Imagine you are a young person living in Boro Gacha village in Bangladesh. Your parents are farmers. Khulna, the nearest city, is many hours away. Until a few years ago, if you wanted to study for an important exam, you had to buy expensive books, travel far to find a good teacher, or just hope for the best.

Now, you pull out your smartphone, open an app, and ask an AI question by question, in your own language, until you understand.

This is not a story about technology. It is a story about one of the oldest human needs: access to knowledge. And for the first time in history, that access is becoming genuinely universal.

From Clay to Conversation: The Longest Revolution

Human beings have been solving the problem of information storage and sharing for about five thousand years. Clay tablets in Mesopotamia kept records for kings and priests. Paper made information portable. The printing press made it mass. Computers made it processable. The internet made it networked. Each step expanded who could access knowledge, and how quickly.

But here is the thing about that ladder: most of the world climbed it rung by rung. Populations in wealthy countries moved from paper to typewriter to desktop computer to laptop to smartphone, carrying the habits of each era forward into the next.

Bangladesh, and much of rural South and Southeast Asia, missed one of those rungs entirely. There was no desktop culture, no keyboard generation, no browser habit built patiently over decades. When the smartphone arrived, it did not arrive as the logical next step in a sequence people had been following. It arrived as a rupture. A device that bypassed everything that came before and handed people something they had never had: a personal, portable, voice-enabled connection to the world’s information, in their own language, for a few rupees a day.

That is not a gap story. That is a leapfrog story. And the consequences are only now becoming visible.

In India, the inflection point came in 2016, when Reliance Jio launched ultra-cheap mobile data and made smartphones accessible to hundreds of millions of people who had never been online before. A farmer in Bihar and a weaver in Madhya Pradesh could suddenly connect to the internet for the cost of a cup of tea. Within a few years, India crossed 958 million active internet users, with 548 million of them living not in cities but in villages and small towns. Rural India is now growing its internet use at four times the pace of urban India.

The smartphone, once a luxury, became infrastructure: as essential and as ordinary as a bicycle or a cooking stove.

The thing those hundreds of millions of new users reached for first was not email, not government portals, not banking apps. It was short-form video: television reimagined for a five-inch screen, in your own language, made by people who looked and sounded like you. As of 2025, 588 million Indians watch short-form video regularly, and rural viewers now outnumber city viewers. After the Indian government banned TikTok in 2020, homegrown platforms like ShareChat and Moj filled the space, with most of their users coming from small towns and villages. People spend between 45 and 60 minutes a day watching these videos. Farmers watch crop tips. Young women watch health tutorials. Students watch coaching classes for competitive exams.

And then, almost before anyone noticed, AI arrived on top of all of this.

What AI Is Actually Doing

According to India’s Internet and Mobile Association, 44% of India’s internet users already use AI-enabled features: voice search, image recognition, chatbots and AI filters. That number is moving fast and moving young.

Take education. In thousands of villages, students are using ChatGPT and similar tools to understand science concepts their teachers cannot always explain, in languages their textbooks do not always use. AI is becoming the personal tutor that only wealthy urban families could previously afford. It does not replace the teacher. It fills the gap when the teacher is absent, undertrained, or simply overwhelmed by a classroom of sixty children.

In healthcare, the gap AI is filling is even more stark. Rural India does not have enough doctors or nurses. Many villages have no health centre at all. AI tools are stepping in as a first point of contact: helping people understand symptoms, offering basic guidance, and directing serious cases toward the right care. A programme called Adi Vaani works specifically for tribal communities, providing health and government information through voice in native languages. For someone who has never had a doctor within easy reach, a voice-based tool that speaks Gondi or Santali is not a novelty. It is a lifeline.

In farming, AI is getting practical in ways that matter directly to household income. Satellite imagery analysed by AI can tell a farmer when to irrigate, which diseases might be developing in a crop, and what market prices look like before the harvest arrives. The reduction in post-harvest losses alone can be transformative for smallholder families operating on margins where a single bad decision means a bad year.

The government is also moving, if unevenly. A chatbot called Jugalbandi, allows anyone with a basic Android phone and a WhatsApp account to ask questions about government programmes in 10 Indian languages and receive answers covering 171 welfare schemes. A farmer does not need to travel to a district office, navigate an English-language website, or know which ministry handles which benefit. They send a voice message. The chatbot replies. That is a quiet but significant shift in the relationship between citizen and state.

Other programmes are targeting specific populations. Kisan e-Mitra helps farmers understand agricultural benefits. Suman Sakhi in Madhya Pradesh connects pregnant women to health guidance through WhatsApp. TriBoT speaks in local tribal dialects including Gondi, so someone asking about land rights receives an answer in their own tongue, not in a language imported by a bureaucracy that was never designed with them in mind.

On the workforce side, the government has announced free AI training for 550,000 Village Level Entrepreneurs operating Common Services Centres across the country, building a grassroots layer of people who can help their communities navigate these tools. Less visibly but perhaps more significantly, an estimated 200,000 rural workers, the majority of them women, are now employed as data annotators: labelling images, transcribing audio, and classifying content that trains the AI models used by technology companies around the world. Rural India is not just consuming AI. It is, quietly, building it.

The Other Side of the Screen

None of this is uncomplicated. The same infrastructure that delivers crop advice can deliver political manipulation. The same platform that connects a woman in a village to a health chatbot also connects her to viral misinformation that can cause real physical harm. And the same AI that is becoming a personal tutor for a curious student in Nagpur may be deepening the gap between that student and one in Delhi who is using AI to turbocharge a career that was already well on its way.

Start with gender. In rural India, the smartphone is often a shared device, and the person who controls it is usually male. Rural women own smartphones at significantly lower rates than men, and those who do access them frequently report needing permission, waiting for a convenient moment, or having their usage questioned even when they are in paid work. AI tools built for voice and local language help at the margins. They do not fix the underlying norm. The device in the household is a proxy for power in the household, and no chatbot changes that without deliberate design and policy attention.

Then there is misinformation. Short-form video is powerful precisely because it bypasses literacy barriers: you do not need to read, and you do not need to search. You simply watch what the algorithm delivers. But that same quality makes it a near-perfect vehicle for false information. A video claiming a dangerous herb cures a serious illness, or that voting for a particular party will cause catastrophe, travels with the same speed and emotional force as one explaining crop rotation. Research from Bangladesh shows that 83% of online adults have been exposed to misinformation and 11 million people say it has caused them or someone they know real mental or physical harm. That is happening now, in a population still in the early smartphone stage, before AI-generated content arrives at scale. When deepfakes and AI-curated disinformation land on top of this infrastructure, the risks multiply significantly.

I was sitting in panel discussions and hallway conversations at the Phoenix Summit on cybersecurity in Dhaka this month, listening to practitioners from across the region, and one observation kept surfacing in different forms. A rural population that consumes AI-curated content without the tools to critically evaluate it is not simply a digital inclusion problem. It is a national security problem. The same infrastructure that brings someone medical advice in their own language can bring them synthetic propaganda calibrated to their fears. These are not separate risks to be managed separately. They are two sides of the same architecture.

The third challenge is perhaps the most structurally difficult. If AI deepens the capability of those who already have capability, the gap widens rather than narrows. Urban, educated, English-speaking users in India are using AI to accelerate careers, investments and business decisions. Rural users accessing AI for voice queries and basic crop tips are gaining real value, but not the same value. Digital access is not the same as digital inclusion. Handing someone a smartphone and declaring the problem solved is the development equivalent of building a road to a village and calling it economic integration.

The Wider Picture: One Phone, Many Chapters

India is the largest and most documented example of this transformation, but it is not the only one.

Across Southeast Asia, a region of 700 million people with a median age under 31 and smartphone penetration above 70%, something similar is unfolding at different speeds and with different textures. Indonesia has the highest worker AI usage rate in the world at 92%, with 18 million businesses actively using AI tools. Vietnam became the first Southeast Asian country to pass a dedicated AI law in March 2026, moving from adoption to governance in a single legislative step. The region’s linguistic diversity, with more than 1,200 languages and dialects across the ASEAN nations, mirrors India’s multilingual challenge and makes the development of vernacular AI tools not a nice-to-have but a fundamental prerequisite for genuine inclusion.

Bangladesh sits at a particularly revealing point in this trajectory. The smartphone revolution there is real and rapid: 73 million Bangladeshis now use smartphones, double the number from just three years ago, and smartphone access has reached parity with television access for the first time. Short-form Bangla video is the dominant content format. TikTok, despite being reported by only 14% of users, accounts for 16% of all mobile data traffic, suggesting a depth of engagement that surface statistics do not capture.

But the AI layer is essentially absent. Bangladesh is roughly where India was in 2018: the smartphone is in the hand, but the intelligence layer has not yet landed on top of it. The contrast with India’s chatbot ecosystem is striking. In Bangladesh, only 15% of people seeking government services go to official digital portals at all. The rest use YouTube and Facebook, social platforms built for advertising reach, not for civic navigation. There is no Jugalbandi equivalent. Government services that India is beginning to make conversationally accessible remain, in Bangladesh, largely unreachable through the device that most people actually own.

And the risks that arrive with the AI layer are not waiting for the opportunity layer to catch up. Misinformation is already causing documented harm at scale. The gender gap is deeper than in India, with 37% of women relying on shared devices and 39% of women with smartphone access reporting they need help to use it. The digital literacy infrastructure that would allow people to critically evaluate AI-generated content is thin. When AI does arrive in Bangladesh at scale, and it will, the question is not whether it will bring opportunity. It is whether the conditions for equitable, safe and critically engaged AI use will be ready when it does.

That question felt very present at the Phoenix Summit. The panels on deepfakes, disinformation and digital trust were not theoretical exercises. They were responses to a landscape already shifting faster than governance frameworks can follow, and the hallway conversations were, if anything, more urgent than the panels themselves.

The Future Is Being Written in a Hundred Languages

Somewhere in human history, a scribe pressed a stylus into wet clay and recorded something worth keeping. The information was real, the intent was real, but the audience was tiny and access was controlled by the few. Every technology since, paper, print, computers, the internet, smartphones, has widened that circle of access. AI, layered on top of the smartphone, may be the most radical widening yet: knowledge that is conversational, personal, local, and available to someone sitting in a village in Odisha or a riverside community in Bangladesh or an island community in Indonesia, in their own language, through a device that costs less than a week’s groceries.

But that widening is not automatic. It does not happen by physics. It happens, or fails to happen, through decisions: about who designs the tools, whose languages get trained into the models, which communities get the literacy programmes, and whether the women in the household are allowed to hold the phone.

India’s villages may well become the most important laboratory for human-AI coexistence of the twenty-first century. What happens there, whether the benefits reach the women still waiting for permission to use the shared phone, whether the AI layer builds bridges or deepens divides, whether governments design services for the people who most need them or for the people who are easiest to reach, will shape how the rest of the developing world thinks about this technology for a generation.

The village has fallen in love with the smartphone. The smartphone has fallen in love with AI.

And in that relationship, written across a billion small screens in a hundred languages, the future of some of the world’s great civilisations is being quietly, urgently and irreversibly composed. One voice query, one short video, one chatbot conversation at a time.

Every bureaucracy dreams of a single source of truth.

A clean database. One number. One record. One neat little profile that says: this person was born here, on this date, to these parents, and should therefore be allowed to go to school, receive healthcare, inherit land, apply for a passport, open a bank account, and eventually argue with government websites like everyone else.

It is a beautiful dream.

The problem is that truth is annoyingly physical. It begins with a real birth, in a real place, involving real people, often in messy circumstances, with tired nurses, paper registers, family politics, weak internet, and someone’s uncle insisting he knows the chairman.

And if the first record is weak, everything built on top of it becomes a very expensive tower of uncertainty.

Take birth registration. In theory, it is the foundation of legal identity. A child is born, the event is verified, the parents are recorded, and the state creates the first official identity record. That record then becomes the basis for school enrolment, health records, social protection, national ID, passport issuance, inheritance claims, and many other things that make a person administratively real.

In practice, many systems are far less elegant. A child may be born in a rural hospital under one authority, at home under another informal pathway, or registered later through a local office based largely on declaration. The hospital may record the birth. The local registrar may record the child. The family may provide the parent details. A village-level actor may facilitate the application. A digital system may eventually produce a certificate. Everyone has a piece of the story, but nobody fully owns the truth.

This is where the comedy begins, and the policy risk hides.

Imagine, years later, a man discovers that three different children have him listed as their father. He is surprised. His wife is more surprised. The database, however, is calm. According to the system, congratulations.

This sounds like a Bollywood inheritance plot, but the underlying problem is serious. If birth event verification is weak, parentage is poorly checked, late registration is easy to manipulate, and multiple agencies hold partial records, the identity system becomes vulnerable to error, fraud, duplication, and exclusion.

The issue is not simply whether a country has a digital birth registration system. A bad paper record is a problem. A bad digital record is a problem with broadband.

Digitisation does not automatically create trust. It can simply move weak evidence faster across government.

The UK Post Office Horizon scandal offers a dark lesson from outside the identity sector. There, a computer accounting system was treated as more reliable than the people using it. When the system showed financial shortfalls, many sub-postmasters were accused of theft or fraud. The institution trusted the database, even when reality was screaming that something was wrong.

The lesson is not “never trust databases.” That would be inconvenient, especially for governments that have already bought the software.

The lesson is this: a source of truth must first be true.

A database is not truth because it is digital. It is not truth because it has a login page, a dashboard, a QR code, or a donor-funded interoperability layer. It becomes trustworthy only when the evidence behind it is reliable, the process is auditable, errors can be corrected, and people can challenge what the system says about them.

India’s Aadhaar experience provides another caution. A foundational identity system can improve service delivery, but when authentication fails, records are not properly linked, or local systems do not reconcile with real life, people can be wrongly excluded from services. In other words, the system may know who you are, but the ration shop may still disagree.

This is why birth registration matters so much. It is not just a certificate. It is the first administrative claim the state makes about a human being: Who was born? When? Where? To whom? On what evidence?

If those questions are answered casually, every downstream system inherits the uncertainty. Schools may register children based on one version of identity. Health systems may hold another. Passport authorities may demand corrections. Social protection programmes may struggle with duplicates. Courts may later face inheritance disputes. And somewhere, a government official may explain, with great confidence, that “the system shows this information.”

That sentence should terrify us.

For countries strengthening digital identity systems, the priority should not be to worship a single database. The priority should be to build a trusted birth event verification chain.

That means linking hospitals, clinics, community health workers, local registrars, and civil registration systems into a coherent pathway. It means recording not only the child, but the evidence behind the record. It means clear responsibility for hospital births, home births, late registration, corrections, parentage disputes, and audit trails. It means designing systems for the real world, where people lose papers, spell names differently, migrate for work, deliver babies at home, and sometimes lie.

A single source of truth is useful only if it is supported by multiple sources of evidence.

Otherwise, the state does not have an identity system. It has a rumour management platform with certificates.

Birth registration should therefore be treated as more than a clerical function. It is the foundation of legal identity, public service access, demographic statistics, child protection, inheritance, migration, and eventually digital government itself.

If the first record is strong, later systems can verify, connect, and serve people better. If the first record is weak, later systems may simply automate confusion.

So yes, every modern government needs an authoritative source of truth for identity.

But before connecting it to health, education, passports, social protection, banking, and digital services, one small question is worth asking:

Is it actually true?

Because otherwise, one day, the database may knock on your door and say:

Congratulations. You are the father.

Scroll long enough and you will meet them all: the 19-year-old founder, the podcast philosopher, the guy in a hoodie explaining how “life is the real classroom.” Universities are framed as optional. Structure is portrayed as weakness. AI is apparently your new teacher. Somewhere in the background, a motivational soundtrack plays.

It is a compelling story. It is also wildly incomplete.

We love the dropout myth because it feels efficient.

Why sit through lectures when experience teaches faster? Why struggle through assignments when failure builds character? Why invest in learning environments when chaos is free?

But here’s the inconvenient detail that never makes it into the reels: learning does not come from failure by itself. Failure is just raw material.

Without reflection, guidance, and feedback, failure mostly produces stress. Or avoidance. Or, if you are lucky, privilege-based recovery. No reflection equals no learning. Just scars.

Life does not come with a curriculum.

life does not sequence difficulty. It does not pause to explain why something went wrong. It does not say, “You were close, try adjusting this.” It simply delivers outcomes. Sometimes generous. Often confusing. Occasionally brutal.

Real learning works differently:

People learn by trying something, reflecting on what happened, extracting meaning, and applying again. They learn best when challenges stretch them slightly beyond their comfort zone, with someone nearby to help. They climb gradually from remembering to understanding to applying, before they ever get to creating.

You do not leap into mastery through repeated faceplants, you climb! Drop someone into deep water without a lifeguard and call it “learning,” and you are not being innovative. You are outsourcing responsibility.

Somewhere along the way, we started romanticizing failure as if it were a teaching assistant.

“Fail fast,” they say.

Sounds brave until you remember that failure has unequal consequences. For some, it is a pivot. For others, it is a lost scholarship, a cancelled visa, or a month of rent they cannot cover. The success stories get podcasts. The rest quietly disappear.

Failure does not educate by default, it selects.

This is where institutions quietly matter, even when they are unfashionable. Not because they are perfect. But because, at their best, they engineer learning environments. They provide structure. They sequence difficulty. They offer feedback. They create spaces where mistakes are safe and recoverable. They allow drafts before verdicts. They surround people with peers who are struggling through the same material.

Universities reduce the cost of learning.

They turn expensive mistakes into guided practice.

If raw failure were enough, we would not bother training pilots in simulators. We would just hand them keys and wish them luck.

We do not. For good reason.

Then comes the latest upgrade: AI will fix everything.

I like technology. I use it daily. But tools are not teachers. Information is not understanding. Speed is not mastery.

AI can answer questions. It cannot notice when your confidence drops. It does not redesign tasks when you are almost there but not quite. It does not create psychological safety. Tools amplify environments. They do not replace them.

If the culture prizes output over understanding, technology accelerates superficiality. If the culture values reflection, feedback, and humane pacing, technology can support growth.

The tool is not the point, the environment is.

The real crisis is not that learning is changing. It is that we keep redesigning learning around efficiency instead of people. Dashboards replace conversations. Automation replaces mentoring. Metrics replace reflection. Everyone is told to adapt faster while the scaffolding quietly disappears. Then we wonder why everyone feels exhausted.

Maybe the future of learning is not about abandoning universities or worshipping disruption. Maybe it is about building spaces where practice is protected, feedback is human, and progress is visible. Where mistakes are treated as data, not identity. Where tools serve development instead of dictating it. Where people are allowed to stretch without being shattered.

Not viral or glamorous; just deliberate, because learning does not come from failure. It comes from practice, feedback, reflection, safety, scaffolding, and time.

Failure can be part of that.

It is not the engine.

Every time someone says “Bangladesh must achieve 100 percent birth registration by 2030,” I imagine a giant scoreboard somewhere where the government is trying to “collect” babies like Pokémon.

It sounds bold. It sounds measurable. It also sounds like something drafted in an air-conditioned conference hall by people who have never witnessed a Bangladeshi family debate (sometimes for weeks ) about what to name their newborn child.

The global CRVS agenda likes to frame birth registration as a state objective. The logic goes:

“Every child must be registered by the government, on time, for national planning and digital governance.”

This is well-intentioned. But it subtly turns citizens into supporting characters in their own story. It treats families as “informants,” parents as “duty bearers,” and newborns as “events that must be captured.”

In reality, families don’t wake up thinking, “Let me satisfy Indicator 1.1.a today.”
They think, “What on earth are we going to name this child?”

The Naming Wars (A CRVS Horror Story)

Let me share a truth every registrar knows: no birth registration system survives contact with actual human behaviour.

Some families spend nine months picking names. Others spend nine months not picking names, and then panic right after delivery.

Then the true drama begins:

• The grandfather insists the baby should be named after his father.
• The grandmother suggests a name that was popular in 1954.
• The father wants something modern.
• The mother wants something meaningful.
• The uncle says, “Why don’t you let me decide? I have a good feeling for names.”

By the time the family reaches the registration desk at a UDC, they present a spelling that no one — including themselves — can reproduce consistently.

Now multiply this by 3 to 4 government systems:
BDRIS, NID, Passport Office, School records.

The result?

One child.
Four identities.
Zero semantic interoperability.

I have seen “Mahin,” “Maheen,” “Mahiin,” and “Mahinn” used for the same child. The CRVS system blames “data quality issues,” but the truth is: the child began life in a naming war.

CRVS Is Not a Process Map. It’s a Human Comedy With Bureaucratic Consequences.

Modern civil registration is often depicted in neat flowcharts.
But real-world CRVS is messy, emotional, and rooted in cultural practices.

Births do not wait for office hours, deaths do not occur in the presence of the official informant, names do not align with rigid data schemas.

And families do not schedule life around process maps.

The problem is not that families are “non-compliant.” The problem is that our view of CRVS was designed backwards.

We start from: “How do we make citizens follow the process?”

Instead of: “How do we design a system that reflects how citizens actually live?”

CRVS in a Digital Age: Less Control, More Humanity

In today’s world, the purpose of CRVS is not to meet a target. The real purpose is to ensure every person can prove who they are so they can access rights, services, entitlements, and protection. It is not about the government “capturing events.” It is about enabling people to belong.

This means reframing CRVS from a top-down administrative mission to a digital identity foundation built around how families truly behave.

A modern CRVS system must embrace the realities of human life. It needs to accommodate messy decision-making, low-connectivity environments, multilingual and low-literacy interfaces, assisted digital pathways, offline workflows, and clear spelling standards. It also needs reliable digital signatures, secure data exchange, and event-driven registries. These features are not required because donors want them. They are required because families want simplicity, dignity, and recognition.

Let the Government Plan. Let Families Name. Let the System Adapt.

If we truly want a world where every child is registered, we must stop imagining CRVS as a “state coverage goal” and start imagining it as a service that wraps around real family life.

The truth is simple: People are not refusing to register babies, they are refusing to wrestle with a system that does not understand their reality.

CRVS is not about control, it is about belonging. And belonging begins with a name chosen through love, conflict, compromise, and sometimes, comedy.

If we design CRVS for that world instead of the world of process maps, everything else will fall into place. Inshallah.

The illusion of certainty

In development, reform, and startup worlds, we often celebrate solutions. Apps, interventions, platforms, and training programs are launched with enthusiasm, each promising to fix something important. Yet many begin with the same fatal flaw: they start with the answer.

People build before they listen, or they listen only to people who make them comfortable. Even when background studies or consultations are conducted, they often confirm what the designers already wanted to believe. The result is predictable. Startup graveyards are filled with apps nobody uses, and projects that teach entrepreneurial skills without improving income.

A moment of clarity

One afternoon, my colleague Rabbany stormed into my room, visibly frustrated but smiling.
“I’m glad our approach doesn’t force a solution onto anyone,” he said. “Nobody does that anymore, or at least almost nobody. Everyone already has a solution before they even understand the problem.”

He was right.

We were in the middle of designing our Digital Co-Design Lab, a process meant to help government teams improve digital public services with citizens, not for them. The whole idea was to make discovery, not prescription, the heart of the work. We wanted to learn what mattered most to citizens before suggesting what to fix or build.

But that is easier said than done. At every step, someone asks, “So, what is your proposed solution?” before deciding whether to join, approve, or fund the work. It feels as if curiosity is not a valid strategy, and people trust only what they can already name.

That is how we ended up with a paradox: to gain permission to explore, you first have to pretend you already know the answer.

The problem with the static plan

Most failures do not happen because the idea was bad. They happen because the idea assumed the world would stand still. In reality, every intervention reshapes the environment it touches. Policies trigger resistance, platforms create new habits, and training changes incentives. Yet many projects move forward with three-year work plans as if certainty can be locked in advance. That is where failure quietly begins.

Learning from effectuation

My entrepreneurship mentor, Wayne Chang, used to say that effectuation is the art of co-creating with uncertainty. His entrepreneurship educations programs wre built on the idea that you start with what you have, who you know, and what you can afford to lose. Then you adapt as the environment responds.

This principle applies far beyond startups. In governance reform, social innovation, and public service design, we need fewer perfectly planned solutions and more small, testable moves that learn as they go.

From answers to questions

Maybe it is time to stop rewarding those who have the right solution and start rewarding those who ask the right questions. Real innovation does not come from knowing the answer in advance. It comes from staying curious enough to notice how the problem keeps evolving.

When digital services are designed to exclude, not empower

I thought the future would be frictionless. Then I tried digitally signing a PDF in Armenia.

Let me paint you a picture.

It’s 2025. Digital transformation is the buzzword in every policy document, investor pitch, and TEDx stage. Governments are promising seamless e-governance. Telecoms are preaching about “secure identities.” And somewhere in Silicon Valley, an engineer is pitching yet another blockchain-based solution to… whatever.

Meanwhile, I’m standing in line at a telecom office in Yerevan, Armenia — holding my expired Mobile ID, a dead SIM card, and a face that screams, “why did I think this would work?”

The mission? To digitally sign a document.

Spoiler alert: It didn’t go well.

Tech that Doesn’t Work for Humans

Let’s break this down.

To remotely sign a simple document, I had to:

1. Replace my invalid SIM card-based Mobile ID, because the one I bought in 2023 was suddenly declared obsolete by my telecom provider.
2. Visit their office in person (so much for digital, right?).
3. Receive a seven-page instruction manual. Yes, seven.
4. Download a government-approved Windows desktop app — no mobile version, naturally.
5. Register on a third-party identity verification system that looks like it time-traveled from 2008.
6. Link both platforms together, assuming you have the right browser, operating system, and a PhD in tech troubleshooting.
7. And finally, if all the stars aligned, apply a digital signature to a PDF document.

All of this… just to digitally sign one file. After looking at seven-page instruction on my desk for a week, I decided to give up. For the record, I have a technology management degree from a world class university.

If that’s digital empowerment, I’d hate to see digital punishment.

The System Isn’t Broken: It Was Built That Way.

While waiting at the service desk, a man in his sixties next to me was pleading with the telecom agent to help him recover his lost PIN and PUK codes.

Why? Because his SIM was locked, and with it, access to all his business contacts.

Did he get help? No.

He got a printed contract clause stating it was “his responsibility,” a QR code to track his grievance, and a suggestion to submit a complaint through the telecom’s online-only portal.

He looked at me and said, “Maybe I should go back to paper phonebooks and landlines.”

Honestly? He’s not wrong.

Digital Services Are Becoming the New Bureaucracy

Here’s the thing.

These systems aren’t failing because of some innocent coding error or a missed deadline. They’re failing because they’re designed without empathy, without user testing, and certainly without the input of everyday citizens.

They’re built by engineers for other engineers, not for people who just want to renew a SIM, sign a form, or talk to a human when something breaks.

We’ve made tech a replacement for trust, not a facilitator of it.

We’ve made it harder to be a citizen in a digital society unless you’re also a developer.

Engineers Aren’t the Enemy, But They Should Study Philosophy!

At this point, I genuinely wonder:

Should engineers be allowed to design citizen-facing systems without understanding human behavior, ethics, or history?

Forget machine learning, maybe we should start with moral reasoning.

Before shipping another portal, app, or tokenized ID, perhaps require every engineering student to pass a course on civic empathy.

Because what we’re dealing with now isn’t just tech failure, it’s social exclusion at scale.

So here’s my question to policymakers, digital architects, and startup founders:

What happens when half the population stops trusting the systems you built?

And do you even care? As long as the analytics dashboard says “user onboarded”?

Governments everywhere are racing to adopt Generative AI. New copilots, chatbots, and content tools promise smarter services, faster processing, and more productive public servants. But there’s a deeper truth many governments are reluctant to confront:

They haven’t yet completed the last transformation yet!

Many services still rely on paper-based approvals, duplicate forms, and legacy systems. What’s called “digital” is often just a PDF uploaded to a website. And now, those same institutions want to leapfrog into AI without ever having built the foundations for automation.

From Paper to PDF Isn’t Transformation

Putting a paper form online is not digital transformation. Nor is emailing scanned documents or downloading Excel files from a government website. Yet for many public institutions, these steps are still presented as progress.

True transformation requires process redesign, inlcuding removing unnecessary approvals, integrating backend systems, and automating repetitive tasks. Without this, digital tools merely replicate outdated workflows, shifting the burden onto citizens to navigate complexity.

The result? Services may appear digital, but they function like manual systems with a digital face. This is not modernization. It’s manumation: the automation of inefficient manual processes.

Research Shows: Technology Isn’t the Problem

A growing body of evidence confirms that the real obstacles lie within institutions and not the tools they use.

A Harvard Business Review article bluntly states: “Digital transformation is not about technology.” It’s about rethinking how work gets done, aligning teams, and shifting mindsets. Technology magnifies existing dysfunctions unless institutions are willing to change how they operate.

McKinsey’s analysis of over 1,000 transformation programs shows that organizations lose up to one-third of expected value before any technology is even deployed. Why? Because they fail to set clear goals, redesign workflows, or build the internal muscle to deliver and sustain change.

Even in the UK, where public sector time-use data shows that 41% of work could theoretically be automated, real-world AI deployments show marginal gains: just 26 minutes saved per worker per day. The tools are powerful. The systems they enter are not ready.

AI Is a Mirror, Not a Fix

The arrival of AI isn’t a shortcut to leapfrog transformation. It’s a stress test for how incomplete the last wave of reform has been.

Plugging advanced tools into broken systems doesn’t solve core problems , it only amplifies them. AI copilots may summarize better or auto-fill more forms, but they still rely on the same fragmented workflows, redundant verifications, and siloed approvals. Without redesign, these tools simply help people navigate bad processes slightly faster.

What Governments Must Fix First

To benefit from AI tomorrow, governments must finish the work they started years ago. That means:

• Redesigning services end-to-end, not just digitizing front ends
• Removing paper-era constraints, redundant steps, and manual dependencies
• Building institutions that can manage change, not just procure tech
• Centering on citizen journeys, not internal convenience

The most transformative public sector examples from Estonia, Singapore, some Indian and Latin American states succeeded not because of any one tool, but because they reimagined how services are delivered from the ground up.

AI Won’t Save You From the Work You Refused to Do

The current AI moment is not an opportunity to skip ahead. It’s a reminder that digital transformation is unfinished , and the hard work of institutional change still lies ahead.

Governments don’t need to wait for AI to start improving. They need to get the basics right: simplify processes, integrate data, empower teams, and design around users. Only then can advanced tools like AI be truly transformative.

Until then, the most advanced algorithms will keep running on top of yesterday’s paperwork.

It was 19997. I was a fresh graduate in a small town of under 20,000 — the kind of place where career options were limited, and ambition often ran headfirst into geography.

My dad, ever the practical man, took me to the local bank. The branch manager was his former student. The ask was simple: an entry-level position, maybe a desk job to get me started.

The manager welcomed us warmly, then asked to speak to me alone.

“What do you want from life?” he asked.

I froze. I hadn’t really thought about it in any meaningful way. I muttered something vague and forgettable. He listened, nodded, and told me his story.

He had started in that very branch after university. Decades passed. He moved through roles — predictable, linear, local. And now, near the end of his career, he was sitting across from me, offering a choice.

“There are two options from here,” he said. “One — take the desk next door. Start small. One day, maybe, you’ll replace me. Or two — turn around, walk back out that door, and run. As far and as fast as you can.”

What I chose is a story for another time.

But that question — “What do you want from life?” — has stuck with me. No one had ever asked me that before — not because I was overlooked, but because in the Soviet Union, that question didn’t exist. You didn’t ask what you wanted. You asked where you were needed. The system placed you. That was the game.

Suddenly, I was supposed to have a vision. A plan. As if decades of centrally planned thinking could be undone with a diploma.

Instead, we were handed the myth of modern career advice:
Follow your dreams.
Pick something you love.
Everything else will follow.

But here’s what no one tells you — that first job offer? That first “safe” choice? It’s not just a beginning. It’s a path. A slope. Sometimes even a trap.

We talk about careers as if they’re expressions of inner truth — like we all have this perfect, pre-formed identity waiting to be unwrapped. But in reality? Most careers are forged like raw material. Not discovered. Not dreamed. Refined.

Imagine a bar of iron — about 1,000 grams. On its own, it’s worth maybe $124.

But shape it into horseshoes, and its value grows. Craft it into sewing needles, and it’s worth tens of thousands. Shape it into watch springs, and now you’re in the millions. Refine it into microchip-grade laser components? That same bar can be worth $15 million.

The value doesn’t come from what the iron is. It comes from what it’s turned into — slowly, precisely, under pressure.

That’s how real careers are built.

The people who thrive aren’t the ones who followed a passion. They’re the ones who learned to shape, sharpen, and reforge themselves. They didn’t wait for clarity. They built skills, took hits, iterated, and adapted — until the world needed what they could offer.

So let me leave you with a question:

Think back to the moment your career trajectory truly began.

Was it a plan? A coincidence? A panic move?
Did it come from passion — or pressure?
Was it a structured decision… or just the best option in a small-town bank?

What shaped that moment most: your own intention — or the world around you?

And if you had the chance to start again, with what you know now — 
would you choose the same door?

Or walk right past it?