When AI started maturing, I thought: this is the moment. The economics would finally work. The disruption was possible. So I went and built it.

Four years later, I understand something I didn’t then - and it’s changed how I think about what we’re actually building, and how big it could become.

I went into building Duckbill with a clear thesis: that AI would finally make low-end disruption possible in this category. Clay Christensen’s framework is useful here - the classic low-end disruptor takes something expensive and exclusive, strips it down, makes it accessible to people the incumbents were ignoring, and eventually climbs upmarket. Personal assistance was a perfect candidate. High price point, wildly fragmented with no vertical integration, essentially zero penetration below a certain income level, and a customer base - working parents, dual-income households, professionals managing more than any one person reasonably should - that was massively underserved and would accept a simpler version of the product if the price was right. AI would compress the economics. We’d build the product. Everyone would get access to something that used to cost $40,000 to $200,000 a year.

That thesis was right. But after four years of actually building it, I’ve come to understand that the real disruption is something larger and more interesting than I originally conceived - and it doesn’t look like what most people building in this space think it looks like.

After four years of operating at the intersection of AI and real-world task completion, I see that AI products that are struggling to convert awareness into adoption are facing two compounding problems. The first is that the ideal surface through which people will actually consume AI remains genuinely unsettled - app, API, voice, agent, something we don’t have a name for yet. The second, and less discussed, is that most AI products stop at the screen regardless of surface. If something can be resolved digitally, it has largely already been optimized - the internet has had thirty years and mobile has had twenty, and the easy software wins are mostly captured. What remains, what still consumes enormous cognitive energy and time for most working people, is everything that requires actual interaction with the world. The phone call that has to happen. The vendor who won’t respond to a form. The appointment that runs through two insurance systems and three calendars. The task that looks simple until a human being on the other end makes it complicated. This is where people’s lives actually break. And this is precisely where most AI products hand the problem back to you.

The last mile - as any operator will tell you - is where most of the value lives and where most products fail to go. What I’ve learned, and what I think the market has not yet fully absorbed, is that bridging that gap doesn’t come from a better model. It comes from years of operational learning about where AI reaches its limit, what judgment looks like at the handoff, and how to build a system that can complete real-world tasks reliably at scale.

And this is a way bigger opportunity than a personal assistant app: once you’ve built the execution layer that makes real-world task completion accessible at scale, you’ve actually solved a different and more fundamental problem. The personal assistant was never the right product - it was a workaround, a bandaid slapped on a world that was fundamentally unreliable and lacked the tools to deal with that unreliability systematically. The wealthy hired people to absorb the chaos so they didn’t have to. What’s being built now isn’t a cheaper version of that bandaid. It’s the infrastructure that addresses the unreliability itself - the system that makes the world more navigable for everyone; a solution that was never in the market because it was never economically or technically possible to address.

This is not a billion-dollar opportunity - it’s a trillion-dollar one. And it’s been borne out of the insane, perhaps ill-advised willingness to operate at the difficult, unsexy, often-humbling intersection of AI and human judgment for years before anyone was writing think pieces about why it mattered.

And it turns out, infrastructure is always worth more than the applications built on top of it. AWS is over half the value of a $2.2 trillion company. Stripe is approaching $200 billion. Neither built the best app; they built the layer everything else runs on. That's what's being built here, for a market orders of magnitude larger: everything that needs to get done in the real world that no software product has ever been able to touch.

I’ve thought a lot about what made that night in Amsterdam so awful. It wasn’t that I was disorganized or unprepared. It was that the world was unreliable in ways I had no tools to address. The plug wasn’t compatible with my pump. Medela’s customer service hours didn’t account for my time zone. I was exhausted, far from home, and completely without leverage. A personal assistant would have helped - would have absorbed the chaos on my behalf, made the calls, sourced a solution. That’s a real service and a real unlock. But what I’ve spent four years building is something more ambitious: not a bandaid for an unreliable world, but infrastructure that makes the world less unreliable in the first place. That’s the disruption I didn’t know I was chasing when I made that promise to myself in Amsterdam. It turns out it’s a much bigger one.

The standard diagnosis is GTM failure. Wrong channel. Bad messaging. Bad onboarding. Every founder I know has pulled those levers. That’s not the problem.

Here’s what I’m actually seeing.

Harvey - arguably the most successful of the AI apps - has contracts with approximately 50% of the Am Law 100. Yet widespread everyday adoption across those firms is still evolving. Think about that. You’ve sold the product to half the most prestigious law firms in the country — top down, institutional buy-in, the hardest part of enterprise sales supposedly done — and people still aren’t delegating their work to it. That’s not a sales problem. That’s a trust problem that no contract can solve.

Consumer AI apps are quietly pivoting to enterprise. Not because enterprise is a better market. Because consumer adoption isn’t happening fast enough and enterprise lets you mandate usage while trust catches up. It’s a retreat dressed up as a strategy.

And the consumer/enterprise divide itself is dissolving. Look at how anyone actually uses AI day to day — professional, personal, professional, semi-professional, personal. Nobody uses it inside the lines. The categories that organized the last decade of software don’t map to how people actually live and work anymore.

What all three of these things have in common: the interface layer hasn’t settled. OpenClaw launched weeks ago. MCPs went from engineer niche to serious industry conversation almost overnight. Whether apps survive or one or two ambient agents become the front door to everything - genuinely unresolved. We are mid-paradigm-shift and the new paradigm hasn’t won yet.

Which means adoption isn’t slow because founders are doing GTM wrong. It’s slow because we’re asking people to change how they interact with the world before we’ve solved how they interact with the internet.

Those two things have to happen in the right order.

Asking someone to let AI schedule their appointments, make their calls, coordinate their errands - or replace their attorney, their accountant, their travel agent, their personal stylist - these aren’t search queries. They’re acts of delegation. They require trust that takes time to form, and a surface through which to extend that trust, that doesn’t fully exist yet.

So if you’re building in this category, stop treating GTM like an execution problem and start treating it like a product problem. Meet people where they actually are - not where the deck says they should be. Watch where trust is forming. Watch which surfaces are winning. The front door to the internet is being built right now and whoever gets that right unlocks the whole category.

The breakout moment will come. Every category that matters has one.

But it won’t come from a better funnel. It’ll come when the interface layer settles and delegation finally feels like the obvious choice rather than the brave one.

Build for that moment. It’s closer than it looks.

Services are a $15 trillion global economy. Bigger than software. Bigger than e-commerce. And there’s never been infrastructure for them.

No Stripe. No Shopify. No operating system layer.

Why? Coordinating humans is fundamentally harder than coordinating data. Every task is different. Every customer has unique needs. Every provider has different capabilities.

Marketplaces gave up on solving the coordination problem. They just matched whoever was available to whatever needed doing. That model has never scaled.

Last week, rentahuman.ai went viral. They showed why the timing is finally right - while demonstrating exactly what won’t work.

What Rentahuman Got Right

The insight was correct: AI can analyze and plan, but execution still requires navigating the real world. Phone trees. Vendor relationships. Negotiations with humans who have discretion.

People will pay for this interface layer. The demand is massive.

Rentahuman went viral because they named what everyone’s thinking: AI needs humans to execute, and that bridge has value.

The market exists. The timing is right.

What They Got Wrong

Rentahuman built it as humans working for AI. Post a task, match with whoever clicks accept, hope it works out.

Same model TaskRabbit and Fancy Hands tried for fifteen years.

Here’s why it doesn’t scale:

You need to negotiate a $3,200 medical bill. You think it’s wrong but insurance bureaucracy is impenetrable.

What rentahuman sees: One task. “Negotiate medical bill.” Match with whoever’s available. Hope they figure it out.

What AI sees when you build it right:

Not one task. Six subtasks:

1. Analyze bill against policy → identify coding error (AI, 30 seconds)

2. Research policy language and precedent (AI, 2 minutes)

3. Navigate phone tree to right department (AI, 5 minutes)

4. Explain the coding error using policy language (human with insurance expertise, 15 minutes)

5. Escalate when rep says no (human, reading the rep’s tone and knowing when to push)

6. Verify outcome → bill reduced to $800 (AI, tracking the actual result)

Each subtask measurable. Each improvable. The human handles the two subtasks that require judgment and relationship management. AI handles everything else and routes to someone who’s proven they excel at insurance escalations.

Rentahuman’s provider: spent 45 minutes on hold doing everything themselves, made $20, problem unsolved, no way to know what went wrong.

Infrastructure provider: spent 15 minutes on the subtasks they’re actually good at, made $60, customer saved $2,400, system learned what works.

This is what AI unlocks. Not replacing humans. Decomposing complex tasks into subtasks where quality is measurable and expertise matters.

Rentahuman treats tasks as black boxes. We break them into components. That’s the difference between a marketplace and infrastructure.

The Infrastructure Model That Scales

AI can finally handle what made services infrastructure impossible: turning idiosyncratic tasks into measurable, improvable subtasks.

Every provider says they’re “good at customer service” or “experienced with admin.” No way to verify. No way to route.

Infrastructure changes this. The system learns:

Sarah excels at insurance escalation subtasks but needs AI support on research. Route accordingly.

This vendor’s phone tree requires approach X. This coding error responds to language Y. This type of negotiation works better with specific precedent cited.

Next time any customer needs an insurance escalation - system knows what works and who’s good at it.

Quality becomes predictable. Providers develop verifiable expertise. Economics work because coordination is efficient, not chaotic.

Why This Captures the Value

Here’s the business model difference.

Rentahuman’s model: Low-margin gig marketplace. High churn both sides. Random quality. Customers use once, leave when disappointed. Race to bottom because providers are interchangeable and tasks are black boxes.

Infrastructure model: Customers depend on it daily because outcomes are reliable. Expand usage across categories because quality is consistent. Providers stay because they’re developing valuable expertise in specific subtask types and getting paid for it. Platform captures value because it decomposes complex human work nobody else can.

This isn’t TaskRabbit with better branding. This is the operating system for services.

When you build infrastructure that decomposes tasks into measurable subtasks:

Better customer retention. Reliable outcomes mean customers stay and expand usage.

Better provider retention. Meaningful work that pays well because verifiable expertise means providers don’t churn.

Better margins. Efficiency from intelligent coordination, not extraction from chaos.

Compounding moats. More data = better decomposition = better routing = better outcomes = more customers = more data.

That’s a trillion-dollar business model.

The Business Model Is the Human Model

Infrastructure that decomposes tasks into measurable subtasks isn’t ethics. It’s the only design that scales.

When you break down “negotiate medical bill” into its components, providers develop real expertise. Not generalists who claim they can do anything. Specialists in insurance escalation subtasks. Vendor negotiation subtasks. Medical scheduling subtasks.

The system recognizes what they’re actually good at. Routes them appropriately. Pays for verifiable expertise.

Service becomes mastery. The best providers get the best work in their specialty. Quality compounds.

That creates the dignity economy - not as charity, but as infrastructure that works. We didn’t build it this way to be humane. We built it this way because unmeasurable quality racing to bottom can’t capture value at scale.

The business model and the human model are identical.

What Happens Next

Services are $15 trillion waiting for infrastructure. AI makes it possible - not by coordinating humans better, but by breaking complex tasks into measurable subtasks for the first time.

Rentahuman proved the market exists and proved what doesn’t work. Tasks as black boxes. Humans working for AI. No decomposition. No quality measurement. That generates headlines, not economics.

We’re building it at Duckbill. Ten million interactions teaching us how to decompose tasks, measure quality, route expertise. That operational data isn’t replicable - it’s learned from actually doing the work.

The company that gets this right doesn’t just capture a market. It creates one.

Services have been waiting for infrastructure. AI makes it possible by turning complex tasks into measurable subtasks. That’s not the ethical choice. That’s the only choice that builds a trillion-dollar business.

Fair question. And it turns out, a number of smart people have been thinking about exactly this.

In 1980, a law professor named Dr. Edgar Cahn proposed something called timebanking in response to Reagan-era cuts to social services. The idea: one hour of your time = one hour of mine. You tutor my kid for an hour, I cook you dinner for an hour.

Andrew Yang has been championing timebanking as the solution to the AI economy since 2018. His diagnosis is spot-on: when labor becomes abundant, people still need meaningful work for purpose and structure, not just income. We need infrastructure to distribute that work.

In 2023, Yang and Stephen Dubner launched a timebanking initiative with a nonprofit model and a physical clubhouse in Baltimore.

But here’s the problem: timebanking has been tried for 40 years. It’s never scaled past 40,000 members globally.

Why not?

The Problem With Rejecting Markets

Cahn’s vision explicitly rejected market-based pricing. He believed markets “devalue everything we define as a human being” - caring, nurturing, the invisible labor that holds communities together. His vision was radical: an economy where everyone’s contribution is valued equally, regardless of skill or education.

It was radical. Even beautiful. Yet the concept has struggled to take off due to a few core problems: Is one hour of cleaning your gutters really equal to one hour of walking my dog? Are there enough people in this community willing to do the harder work? And how does the community ensure accountability and quality control?

Well, for all their faults, markets have gotten some things really right. They clear supply and demand better than anything else humans have invented through market infrastructure that efficiently prices, matches supply and demand, and builds self-sustaining accountability mechanisms.

The question isn’t whether to use markets. It’s whether markets can be designed to respect human dignity while actually working at scale.

What Actually Works: Markets With Multiple Currencies

Here’s how it works at Duckbill:

You need someone to negotiate a vendor refund. Market rate: $30/hour.

If you’re on the upward curve of the K, you pay $30 in cash.

If you’re on the downward curve and can’t afford cash, you pay in time valued at market rates. Your project management skills are worth $30/hour? Trade one hour. Worth $20/hour? Trade 1.5 hours.

Same market. Two currencies. Both priced honestly.

It’s not charity. It’s exchange for the AI era. And it scales.

The Path Forward

The dual-currency model solves the affordability problem. But here’s the harder question: why hasn’t anyone built this before?

The answer is AI.

Over the past four years, Duckbill has completed over 10 million real-world interactions. Those interactions have allowed us to train our AI for the unpredictability of the real world. To break tasks down into chunks that humans can do with minimal reskilling and maximum impact. To turn every need into something that an F1 pit crew can handle - swiftly, effectively, with dignity.

And it has given us the world’s largest database of unmet supply and demand. These are tasks people desperately want done but have no market for: negotiating refunds, planning events, coordinating logistics.

No one has built a market to aggregate, price, deploy, and quality control these services.

That is, until now. Duckbill isn’t just facilitating exchanges. We’re mining for work that doesn’t exist yet and building markets around it. With professional pricing. Real liquidity. Quality infrastructure.

Yang’s nonprofit model can’t scale to meet 20% unemployment. It can’t build the professional infrastructure - pricing, matching, quality control - that makes markets actually work.

When AI displaces jobs at the scale Yang predicts, we’re going to need solutions that can handle that scale.

Yang saw the problem. He’s building a community center to solve it.

We’re building the market that works for both sides of the K.

And yet, when I tell people what we do, I watch their faces change.

Not excitement. Something closer to grief.

I know what they’re thinking: What happens when nothing I do is necessary anymore?

And I see this resistance in our own data.

Some of our users want us to automate everything. They’re our power users. They never want to think about logistics again. They want full cognitive liberation.

But other users draw hard lines:

“Can you use a human for the phone calls? I don’t want AI speaking for me.”

“I’m fine with AI for admin, but my kid’s tutor needs to be human.”

“Can you use a human for the sensitive stuff? I don't want AI handling that.”

For two years, I viewed these users as late adopters. People who just needed more time with the product.

Now I realize: They’re showing me the market.

They’re saying: I understand AI is faster and cheaper. I still want the choice to use a human.

And they’re willing to pay for it.

That willingness matters - because it intersects with something much bigger.

The Inversion

Here’s my prediction for 2035:

As AI replaces jobs, demand for meaningful work explodes.

Not demand for income (we’ll save that for another post).

Demand for purpose. For identity. For the feeling that your effort matters.

And here’s what nobody’s talking about: When you combine desperate demand for meaningful work with customers willing to pay premiums for human services, you get a massive market.

When people desperately need work (not for money, but for meaning), and wealthy people are willing to pay premiums for human services, you get an economy where meaningful work becomes the luxury good.

The question is: Do we build infrastructure where that work has dignity? Or do we recreate a servant class?

The good news: we have blueprints for building for dignity.

The Blueprints

Etsy: The Structural Blueprint for Dignity

In the 2000s, global manufacturing made craft goods abundant and cheap. Any artisan competing on efficiency was dead.

But Etsy built infrastructure that let craftspeople maintain dignity anyway:

• Portfolio diversification: You don’t work for one gallery owner. You have 500 customers buying your jewelry, pottery, prints.

• Portable reputation: Your reviews travel with you. You’re not starting from zero if you leave.

• Direct relationships: You own your customer base. Platform just facilitates.

• Low take-rate: ~6.5% vs. employment overhead. You capture the value you create.

Result: Even when supply is infinite, individual craftspeople maintain autonomy and pricing power.

The key insight: Abundant supply doesn’t require servitude if you build the right infrastructure.

Japan: The Cultural Blueprint for Service as Mastery

In Japan, service work isn’t subservient - it’s mastery.

Omotenashi (hospitality) is a lifelong craft. A sushi chef trains for decades. A hotel concierge perfects the greeting. A barista studies the pour.

These aren’t “service jobs” in the American sense (low-status, anyone can do it, you’re lucky to have work).

They’re crafts. You practice for years. Excellence brings status. It’s not “I serve people” - it’s “I’ve perfected this.”

The key insight: Whether service work is high-status or low-status is a cultural choice, not an economic necessity.

What Happens Next

Now combine these insights with the AI future:

As AI replaces cognitive jobs, two things happen simultaneously:

1. People desperately need meaningful work
For identity. For the feeling that they contribute something that matters. The demand for work - work that feels real, that involves human connection, that can’t be automated away - becomes enormous.

2. People pay a premium for human services
Just like organic food, craft beer, artisanal coffee - when the efficient version exists, people pay more for the human version. Not because it’s better, but because it’s meaningful.

The market for human-centric services explodes: human customer service, human tutors, human schedulers, human coaches, human researchers.

But here’s the crucial question:

Do those service providers have Etsy’s structure + Japan’s culture?

Or do they become a new servant class - grateful for work, dependent on patrons, no autonomy, stuck forever?

The Infrastructure We Need to Build

I sit on the board of Boston Beer Company, the company that created the American craft beer market.

Here’s what Jim Koch figured out in 1984: You can’t out-efficiency Budweiser. But you can build a $26 billion market for people who’ll pay 3x for beer made by humans in small batches.

The premium isn’t for quality. It’s for story. For meaning. For choosing something that isn’t algorithmically optimized.

The same pattern will happen with human services in 2035.

People will pay premiums for “human customer service,” “human tutors,” “100% human coordination.”

The question is whether the humans providing those services have:

From Etsy:

• Many clients, not one patron

• Portable reputation across platforms

• Ownership of customer relationships

• Low platform take-rates so they capture value

From Japan:

• Cultural framing of service as mastery, not servitude

• Certification systems that create status hierarchies (Level 1 → Level 3 mastery)

• Community recognition for excellence

• Identity from “I’m excellent at X” not “I work for Y”

This Is What We’re Building

At Duckbill, we started by combining AI + humans to tackle your personal admin. But what we’re really building is infrastructure for the dignity economy.

Not just a marketplace - the foundational layer that makes dignified service work possible at scale:

• Reputation systems that are portable across categories

• Certification infrastructure that creates verifiable skill hierarchies

• Tools for managing 100+ client relationships (impossible to do manually)

• Dynamic pricing based on mastery level and demand

• Built-in pathways from service provider → team lead → owner

When a company wants to offer “premium human customer service” in 2035, they won’t build their own infrastructure. They’ll plug into the dignity economy layer.

When a worker wants to transition from employment to autonomy, they won’t figure out the mechanics themselves. They’ll use the infrastructure we built.

Etsy proved the structure works for crafts.
Japan proved the culture works for services.
We’re building the infrastructure that combines them - for the economy that’s coming whether we’re ready or not.

The Stakes

By 2035, meaningful work is the luxury good.

Not because work is scarce. Because purpose is scarce when nothing you do is necessary.

The market for human services will be massive. The wealthy will pay premiums. The demand from people who need work will be overwhelming.

The only question: Do we build infrastructure that creates artisans or servants?

Etsy proved the structure works for crafts.

Japan proved the culture works for services.

Now we just have to combine them - before the market does it for us in the worst possible way.

By 2035:
Meaningful work becomes luxury.
Human services become premium.
The infrastructure we build today determines whether service providers are autonomous artisans or grateful servants.

I’m betting we build the former.

Not because it’s inevitable - but because we’re building it, based on these blueprints.

Not slow it down. Not regulate it better. Stop it entirely.

That statistic, from a new HarrisX/Merit America poll of 3,048 workers earning under $50k, should terrify anyone who cares about American competitiveness. Because this isn’t happening in Germany. It’s not happening in the Netherlands or South Korea or Singapore.

It’s happening here. And it’s going to cost us.

We Lead the World in AI Anxiety

Recent Pew Research data confirms what the HarrisX poll suggests: Americans are more afraid of AI than workers in virtually any other developed nation. About half of US adults are more concerned than excited about AI - among the highest rates globally.

Meanwhile, workers in Germany, the Netherlands, and Nordic countries are significantly more optimistic about the same technology.

Same technology. Different response. Why?

Here’s what everyone gets wrong: they think this is about education, or tech literacy, or cultural attitudes toward innovation.

It’s not.

Americans enthusiastically adopted smartphones, streaming, e-commerce, social media. We’re not Luddites. We’re not technophobic.

The difference is what happens when you lose your job.

When a German worker loses their job to automation, they lose their job.

When an American worker loses their job to automation, they lose their health insurance, their retirement contributions, their housing stability, their kids’ healthcare, their entire economic security infrastructure.

The variable isn’t technology. It’s infrastructure.

You Can’t Train Your Way Out of Structural Precarity

Here’s where the “upskilling” narrative completely falls apart.

The HarrisX data shows:

• 86% would participate in free AI training

• 76% say it would make them more confident

• Yet 56% still think government isn’t prepared to handle AI

Workers aren’t saying “I can’t learn.” They’re saying “even if I learn, I have no economic security during transition, no guarantee the new jobs will exist, and no assurance my new skills won’t also be automated in three years.”

Training addresses capability. The actual problem is consequence.

Rebecca Taber Staehelin from Merit America calls for “a DARPA-like approach to preparing for the future - not a Katrina-like response once economic change has come.”

She’s right about the urgency. Wrong about the solution.

DARPA built technology infrastructure. What we need is economic infrastructure. Training programs - even brilliant ones - don’t answer the question that’s actually keeping workers up at night: “What happens to me?”

The Trust Collapse That’s Breaking Deployment

64% of lower-wage workers don’t believe Silicon Valley cares about AI’s impact on ordinary citizens.

Only 40% think Democrats are handling AI preparation well. Only 30% think Republicans are.

This isn’t a sentiment problem. This is a deployment crisis.

When workers don’t trust tech companies or either political party to protect their interests, resistance becomes the only leverage they have. That resistance shows up as:

• Slower implementation timelines

• Political pressure for restrictive regulation

• Public backlash constraining product development

• Talent concerns about ethics creating recruitment problems

And here’s what should terrify American business leaders: our competitors don’t have this problem.

How the Infrastructure Gap Becomes a Competitive Gap

European and Asian companies can move faster on AI adoption because their workforces aren’t fighting for survival.

A worker in Denmark who loses their job to AI still has:

• Healthcare

• Housing support

• Retraining with income replacement

• Active labor market programs to help transition

They can focus on learning and productivity. American workers are focused on survival.

This creates a compounding advantage. Countries with stronger safety nets can:

• Deploy AI faster (less workforce resistance)

• Experiment more boldly (workers can afford to fail)

• Capture productivity gains earlier (less friction)

• Build public support (workers see opportunity, not threat)

We’re not losing the AI race because our technology is worse. We’re losing because our social infrastructure makes deployment harder, slower, and more contentious.

We’re creating the worst possible scenario: slow enough to fall behind, chaotic enough to prevent thoughtful governance.

The Policy We Need Won’t Come From Government

Trump’s AI action plan “encourages” reskilling. No large-scale federal program. No funding mechanism. Just encouragement.

Individual states are doing some partnerships and grants. It’s patchwork at best.

The infrastructure we actually need - portable benefits, wage insurance, healthcare decoupled from employment, income support during transition - isn’t coming from Washington anytime soon.

Which means the solution has to come from companies.

Not because we’re charitable. Because it’s strategic.

The Business Model That Wins

The companies that dominate the next decade won’t be the ones most efficiently eliminating humans. They’ll be the ones building AI-human hybrid models that make human work more valuable.

This isn’t just ethics. It’s deployment advantage.

When you build AI that augments rather than replaces:

• Adoption happens faster (workers see opportunity, not threat)

• Outcomes are better (human judgment handles edge cases and relationships)

• Scaling is sustainable (you’re not fighting regulatory backlash)

• Moats are deeper (pure automation can’t replicate human-AI collaboration)

At Duckbill, we combine AI with human workers for real-world tasks - phone calls, appointments, administrative coordination. Our model creates income for workers rather than destroying it.

The results: 90%+ retention after month 3, 72% organic referrals, and 56% WAU/MAU.

People don’t just want AI to work. They want to work with AI.

That’s not an anomaly. It’s a signal about what sustainable deployment actually looks like.

What This Means for American Competitiveness

Lower-wage workers have looked at the future being offered and decided that stopping AI entirely is preferable to economic catastrophe.

That should clarify what’s at stake.

We can keep pushing a replacement narrative that creates ever-greater resistance, slower deployment, and political crisis.

Or we can build augmentation models that address the structural anxiety directly—creating the economic security that makes workers willing to embrace change rather than fight it.

The countries that will lead in AI aren’t the ones with the best algorithms. They’re the ones whose workers can afford to adopt them.

Right now, American workers can’t.

And until that changes - either through policy infrastructure that won’t come, or business models that actually solve for human economic security - we’re going to keep falling behind competitors who figured out that you can’t deploy revolutionary technology on top of a collapsing social contract.

The AI anxiety gap isn’t a PR problem.

It’s a competitive crisis.

And it’s going to determine who wins the next decade.

How are you thinking about this in your organization? Are you designing for replacement or augmentation - and how is worker response affecting your timeline?

MJ is CEO and co-founder of Duckbill, building AI-human infrastructure for real-world tasks. Previously COO at Oscar Health, Regional GM for US & Canada at Uber, and Senior Policy Advisor at the US Treasury.

Subscribe now