The Unbundling of the Corporation: From Pigeons to Platforms

It’s a lazy Sunday afternoon and I’m reading yet another doom and gloom article about ‘the permanent underclass’ that is going round tech circles in the US. The theory goes that AGI will introduce an era where capital is king as ‘AI labour’ will scale in line with how much money you have to deploy on compute power. At that point everyone is ‘frozen’ at their level of social standing and social mobility is no more. 

I’m not so sure.

I then started watching an interview with Amjad Masad of Replit describing employees as a bug not a feature of capitalism. I thought ‘Oh no, here we go again…’ but the rabbit hole that Amjad led me down was not one of a permanent underclass but of a massive unlock of the power of the corporation.

The typical reader of this blog (c.58% according to LinkedIn) works for a company of over 10,000 employees. So most corporate readers already feel some sort of coordination tax acutely. Studies show knowledge workers spend 50–60% of their time on “work about work” - meetings, emails and admin tasks - instead of creating value. In one survey, senior managers reported 78% of their day was consumed by coordination, not productivity. The maths explains why. Every new employee adds exponentially more communication channels: a 10‑person team has 45 possible links; a 20‑person team has 190. In other words, adding staff adds even more overhead, not just linear capacity. As Fred Brooks famously noted, “adding manpower to a late software project makes it later” - and the same goes for any business.

This “coordination tax” is real. A Harvard Business Review found managers spend 23 hours a week in meetings - double the time in the 1960s - and a Doodle survey estimated pointless meetings waste $541 billion of US corporate payroll each year. If that number doesn’t terrify you, it should. The more layers, the more planners and check‑ins we hire, the slower we move. In fact, organisational researchers call this the FTE Paradox: as AI makes each worker vastly more productive (10× or more), simply adding more people yields diminishing - even negative - returns. We’re essentially trying to untangle Christmas lights by adding more tangled strands.

With that, let's dive in.

Small Teams, Micro-Enterprises and Dunbar’s Number

Tech firms have already discovered a better way: stay small. Netflix didn’t grow by building a bigger monolith of engineers – it broke its system into 1,000+ microservices across many domains, each a focused team owning a specific function. Amazon formalised Jeff Bezos’s Two-Pizza Team rule: every team should be as small as can be fed by two pizzas (roughly 5–8 people). These rules weren’t about pizza or cost savings - they acknowledge a simple fact: human brains don’t scale, so neither can workgroups. Even the Chinese appliance maker Haier went extreme, carving its 80,000+ employees into 4,000 independent micro‑enterprises, each with its own P&L.

Why 5–8 people or 150? Anthropology provides the clue: Dunbar’s Number. Researcher Robin Dunbar found that humans can maintain only about 150 stable social relationships. Above that, we lose cohesion and agility. (A recent review of Dunbar’s work suggests even a 4–500 range, but the core insight holds: there’s a cognitive cliff.) In practical terms, once your org grows beyond a few hundred (or thousand) people, natural trust breaks down and you need complex hierarchies and processes to coordinate. That’s exactly when companies slow to a crawl.

Yet most corporations stubbornly remain monolithic on paper. We still have software “microservices” (supported by subteams, Agile squads) running on mainframe org charts. We’ve all seen it first‑hand: hundreds of experts locked in rigid hierarchies, each clamoring for attention through endless meetings and status reports. We intuitively know small, autonomous teams win the race, yet we cling to the old model of scale.

AI: Superpower and Snare

Enter AI, which supercharges these dynamics. One insightful colleague recently described AI as a fertiliser - it turbocharges growth whether you are a needed plant or an unwanted weed. On one hand, every knowledge worker can become 5-10x more effective: studies show 40% faster performance on tasks with AI copilots. That means a single person with a half‑decent AI agent can do the work of many. But there’s a catch: AI also escalates coordination costs. Sure, prompts and generative tools automates tasks, but they introduce new hidden work - prompt engineering, model tuning, data wrangling, integration headaches and human‑AI oversight. Your team might crank out work faster, but now you have to manage model drift, endless API changes, and more cross‑checking.

The net effect? You shrink the unit of work into 1-person microservices, but every microservice now needs orchestration. It’s like having thousands of tiny bots - each one brilliant at one thing - all needing a director to keep them from stepping on each other’s toes. If we don’t redesign the company, all that extra AI-generated output just floods through the same clogged communication pipes. As one recent Fed study put it, AI “jaggedly” amplifies output but doesn’t solve the fundamental co‑ordination problem.

In short, AI widens the performance gap between well-structured teams and bulky bureaucracies. A nimble company that treats people and AI as modular components will capture those productivity gains. A Titanic-sized corporation that piles AI on top of its existing HR mess will simply swamp itself in more chatter and approval chains.

The People-API, Platform Model

So if more coordination is poison, the solution is less coordination – not more management. That means unbundling the firm the way tech unbundled code: core + plugins. Imagine an organisation structured like a platform or marketplace. At the center is a core team of up to a few thousand people – your strategic thinkers, architects, product owners and culture carriers. Around that core is a dynamic ecosystem of specialists (what I call people APIs): on‑demand sales experts, CRM agents, finance microsquads, legal pods, etc. These external providers are themselves small, AI‑augmented firms that plug in to deliver exactly what’s needed, then disconnect.

This is no fantasy: the pieces are falling into place. For example, rather than building its own massive sales force, a company might call youratlas.com - an AI‑driven sales consulting platform - to handle quotas and leads (I have no affiliation). Customer support could be handed off to getrevio.com (ditto!), a chat‑based CRM startup that already “talks” to customers for dozens of clients. Marketing, HR, analytics, you name it: each domain has (or will have) its “Uber for [that function]” startup. You simply subscribe or contract with these niche players.

Research backs this micro‑business revolution. A 2024 study calls it the era of hyper-specialised micro‑businesses: tiny teams or solo experts using AI to deliver enterprise-grade work. These micro‑firms combine the agility of freelancers with the depth of a boutique consultancy. A firm no longer hires 50 data scientists; it taps an on‑demand network of AI‑augmented data strategists. In this model you get zero coordination overhead beyond defining outcomes, because each micro-business takes full ownership of its deliverables.

The platform organisation also relies on new trust networks rather than hierarchy. Reputation systems (like LinkedIn endorsements on steroids) vet talent. Teams form and dissolve per project, not by management decree. Success is measured by deliverables, not hours at the office. In effect, every role becomes a bundle of skills-based APIs: you call it with a brief and budget, it returns a solution.

This isn’t pie-in-the-sky. Microsoft’s 2025 “Frontier Firm” vision sees exactly this: companies built around on-demand intelligence with “hybrid” teams of humans + AI agents. They predict that within 2–5 years every organisation will be on the journey to becoming a Frontier Firm. In that model, expertise is just an API call away - whether human or machine - and businesses scale by spinning up the right micro‑services for each new challenge.

Legacy vs Futuristic Firms: Diseconomies of Scale

Contrast this with the old model. Today’s large corporations are relics of the industrial era: built for factory lines and predictable processes. They treat employees as fixed assets. But in a digital economy, talent is a fluid good. Every full-time hire carries hidden costs: underused specialised skills, layers of managers, HR bureaucracy, compliance and inertia. Innovation grinds to a halt under annual reviews and hierarchical approvals.

As the FTE‑Paradox research shows, stacking those costs on top of AI’s output simply erodes the gains. In effect, big legacy firms face diseconomies of scale: the bigger you get, the less each person is worth. The agile “center + people‑API” firms, by contrast, convert scale into modular leverage – assembling only what they need. (It’s telling that some giants like Amazon have even rebuilt parts of their stack into “modular monoliths” to simplify coordination.)

If your company is still organised like Bell Labs or GM, beware. Every bureaucrat or obsolete team you keep is a festering coordination cost. While rivals stitch together best‑of‑breed talent networks on the fly, you’ll be trudging through another month of all‑hands meetings.

Embrace the Platform or Be Left Behind

I won’t lie - this is a radical hypothesis. It flies in the face of our default assumptions about “real work” and “real employees.” It means imagining a sales force that’s in-part outsourced, or a development lab that’s partly freelance and AI. What if your next VP is not a person, but a platform?

The early adopters (think Haier and the Microsoft Frontier Firm) are already pointing the way. In five years, we may look back and wonder why we ever thought 20,000 FTEs would outperform 3,000 core employees plus a swarm of AI‑powered on‑demand talent. The evidence - and biology - suggests smaller, modular teams will win. The choice for leaders is stark: retool your org for a “people API” world, or watch as your bureaucracy cannibalises the productivity dividend of AI.

After all, would you rather add another level of management, or call an API for a problem solver?

The Bottom Line

The age of the corporate monolith might be over. Not because it's trendy to say so, but because the maths doesn't work anymore. Coordination costs scale exponentially. Human cognitive capacity doesn't scale at all. And AI makes both problems worse, not better.

The winners in the next decade won't be the companies that coordinate better. They'll be the ones who've architected away the need for coordination. They'll look less like traditional corporations and more like ecosystems of micro-businesses.

The unbundling has already begun. The AI-enabled services are growing daily. The infrastructure is ready. The only question is whether your organisation is. So what about the AI underclass? As ever, I’m optimistic. The opportunity for employees is an AI-powered skills-based lens. Evolving yourself into a hyper-niche AI superpowered on-demand skill-API (it’s a bit of a mouthful - I’ll have to work on the branding) means the opportunities for people are going to grow not stagnate.

Until next time, you'll find me at dawn, decomposing my own work into microservices and wondering why it took us this long to apply software architecture principles to human organisations...

Resources & Further Reading

Primary Sources:

• Matthew Berman interview with Amjad Masad: Vibe Coding, Platform Risk, Agentic Future, Permanent Underclass, and more! (YouTube, September 2025)

• Boundaryless: "The Trilemma of Organizational Unbundling" (2023)

• Brooks, F. "The Mythical Man-Month" (1975)

• Organization Science: "Coordination in Dynamic Teams: Investigating a Learning-Productivity Trade-Off" (2024)

Contemporary Research:

• MIT/Harvard Studies on Team Size and Productivity (2023-2025)

• Dunbar's Number revisited - Royal Society (2021)

• Microservices trends and adoption - Multiple industry reports (2025)

Case Studies:

• Amazon's Two-Pizza Teams and move back to monoliths

• Haier's 4,000 micro-enterprises model

• Netflix's microservices architecture