The Parasite Layer
On startups, bureaucracy, and the systems that feed on themselves
I once sat behind a procurement desk in a government body. Not for long, but long enough to learn something that changed how I see every system I’ve ever touched since.
We didn’t care if something cost $300 or $3,000. In fact, we preferred the $3,000 option. Not because it was better. Because it was easier - the supplier is happy, things get done easily, we don’t care. Spending right under the tender threshold also meant no paperwork, no approvals, no questions. The system had been designed to prevent waste, and the system itself had become the waste. Nobody was corrupt. Nobody needed to be. The structure did the work for them.
I think about this every time I look at a new SaaS product. It’s a stretch, I know but bear with me.
The solution looking for a problem
The Silicon Valley startup model has a founding mythology that goes something like this: build something elegant, find product-market fit, scale it to everyone. The implicit assumption is that if the technology is good enough, the problem will reveal itself. Or better yet, you can convince enough people that the problem exists.
This is backwards. And everyone in the industry knows it’s backwards, but the incentive structure rewards it anyway. A founder doesn’t need to solve a real problem. They need to convince a venture capitalist that the problem is real enough and the market large enough to justify the bet. The VC doesn’t need the product to work forever. They need it to work long enough to reach the next funding round or an acquisition. The company that acquires it doesn’t need it to integrate well. They need to show their board that they’re “innovating.” At every step, the incentive is to perform value, not to create it.
The result is an ecosystem where thousands of startups are born every year, most of them building variations of the same thing, competing not on whether they solve a genuine problem but on who can acquire customers faster with cheaper capital. The founders themselves fall into a narrow range. Some are genuinely capable builders who see a real gap. But a growing number are simply people who’ve looked at the math of the corporate ladder, decided the odds are better in venture-backed entrepreneurship, and figured out that the game rewards conviction more than competence. They are optimising for exit, not for impact. Build fast. Raise money. Get acquired or IPO. The product is secondary to the financial engineering around it.
And then there’s a harder truth that the ecosystem doesn’t like to discuss. Some of these founders are simply psychopathic extractive people. They don’t want to build something that lasts. They just want to be able to sell their company and go on to building a new thing to extract more from others. Other founders are genuinely jaded with the “normal” way of life and hope for a way to escape the death sentence of a working life, that they may fast track their way to retirement (I’m a start-up founder too but don’t ask me which am I). Benevolent or malevolent, fact is that the system is incentivising solution creation, and yet these solutions create more problems than they fix.
The SaaS bloat machine
But the real damage is the ones that succeed just enough to get embedded.
Think about what happens when a mid-sized company decides it needs a CRM. They evaluate options, pick one, and sign an annual contract. Immediately they need someone to administer it. Not because the software is bad, but because it was designed for a thousand different use cases across a thousand different industries, and the company needs to configure it for their specific one. So they hire an admin, or they pay a consultant. Then they realise the CRM doesn’t talk to their invoicing software, so they buy an integration tool, or they hire a developer to build a custom connection. Then someone in marketing wants analytics, but the CRM’s reporting isn’t quite right for their needs, so they buy a business intelligence platform and hire an analyst to run it. Then they need a project management tool to coordinate the work being generated by insights from the BI platform that pulls from the CRM that required the integration that needed the admin.
Every layer generates the next layer. The company started with a simple need: keep track of our customers. Three years later, they’re running six platforms, employing four people whose entire jobs exist only to manage software, and paying a combined six figures annually in subscription fees for tools that each do about 30% of what the company actually needs and 70% of what they’ll never touch. Multiply this across accounting software, HR platforms, supply chain management, internal communications, document management, and compliance tracking, and you begin to see the scale of the problem. A company that should be lean is now carrying the operational weight of a small bureaucracy, not because it chose to, but because the software ecosystem is designed to compound.
Salesforce alone generates an entire job category. There are people whose career is “Salesforce Administrator.” Not salespeople. Not strategists. People whose full-time work is maintaining a tool that was supposed to make sales easier. The tool became the job. The tail wags the dog.
Billions have been poured into building these SaaS companies. And then billions more have been spent because of them. Not on the products themselves, but on the infrastructure of humans and systems required to make the products function in a real-world context they were never specifically designed for. The industry created to reduce inefficiency has become one of the great generators of inefficiency in the modern economy.
The market feels this. Investors are tired. Operators are tired. Nobody wants another dashboard. And yet the founders keep coming, because the model still rewards the attempt more than the outcome.
Bureaucracy as organism
This is not a new pattern but a fundamental issue with the idea of systems.
Every government begins lean. A few people, a clear mandate, direct accountability. A city needs water, so it builds an agency to manage the water supply. The agency works. It grows. It needs an oversight board. The oversight board needs administrative support. The administrative support needs a procurement process. The procurement process needs compliance officers. A generation later, there are more people managing the system than managing the water.
This is not cynicism. This is how organisations behave in the absence of constraint. Growth is the default state of any system that doesn’t face regular pressure to justify its own existence. And government is the purest example because the feedback loops that constrain private companies, customer churn, revenue pressure, competitive displacement, don’t apply in the same way. A government agency that performs poorly doesn’t lose its customers. It requests a larger budget.
Here is the perverse thing. Government spending increases GDP. Remember the GDP equation: GDP = C + I + G + (X-M). When the government hires a person, that salary shows up as economic output. When it buys a $3,000 office chair instead of a $300 one, the additional $2,700 registers as growth. A government that spends more on itself appears, by the metrics we use to measure economic health, to be presiding over a growing economy. Which means there is a built-in incentive to keep spending. The moral hazard is structural. It is not a bug and feature of how we’ve chosen to measure success.
And within those layers, corruption lives in the gaps. In the procurement officer who picks the expensive option because the process for the cheap option is more work. In the contractor who knows exactly where the threshold sits and prices accordingly. In the middle manager who has no skin in the outcome and every incentive to keep things exactly as complex as they are, because simplification is a threat to their role’s existence.
DOGE and the broader conversation around US government bloat have made this visible in ways that are difficult to ignore. The US federal budget deficit crossed $1.8 trillion in recent years, and while much of that is driven by entitlement spending and debt service, a meaningful portion reflects the accumulated weight of agencies, programs, and processes that have grown beyond their original mandate. The pattern is universal. Singapore has its own version, albeit more controlled, but present. Every large corporation has it. Anywhere a system grows unchecked by consequence, the organism feeds itself.
The case for systems, and the case against them
I am not arguing that systems are the enemy.
A good system is a defence against the worst of human nature. Procedural controls exist because without them, that one bad actor, that one moment of greed or negligence, slips through and causes damage that takes years to repair. The reason we have procurement rules is because without them, a single corrupt official can funnel millions to a friend’s company and no one catches it until the building collapses or the medicine doesn’t work. Systems encode institutional memory. They protect the organisation from the randomness of individual character.
But there is a spectrum. On one end, a well-designed system creates accountability without paralysis. On the other end, too many systems remove all skin in the game. Nobody is responsible because everyone is following the process. The process itself becomes the purpose. Individual judgment atrophies because the system doesn’t require it, and eventually doesn’t permit it. The person at the desk stops asking whether the decision makes sense and starts asking whether the form has been filled out correctly. The body grows organs it doesn’t need, and the organs consume resources meant for the mission, and the larger the body, the longer the processes and the more susceptible to incompetence and corruption.
This is true for governments and for-profit entities alike. The bloat doesn’t discriminate. A Fortune 500 company with 14 layers of management and a compliance department that takes six months to approve a vendor contract is suffering from the same disease as a federal agency with overlapping mandates and no sunset clauses. The organism protects itself at the expense of its stated purpose.
Yet knowing this changes nothing
The frustrating thing is that the simplification isn’t hard to imagine. That’s what makes the problem so difficult to accept.
Take the legal system which I am rather familiar having been trained in it.
The current system has judges, lawyers, juries in some jurisdictions, clerks, paralegals, administrators, and legalese and a procedural architecture so dense that a person with a legitimate grievance cannot navigate it without paying someone thousands of dollars to translate their own experience into language the court will accept. A straightforward contract dispute can take two years and cost more in legal fees than the amount in contention. A family going through a custody arrangement needs to retain lawyers on both sides, each billing hundreds per hour, to resolve something that is fundamentally a human question about what’s best for a child.
The legal profession will argue this complexity is necessary. That the stakes are too high for simplification. That the adversarial process, where each side presents its strongest case through a trained advocate, produces better outcomes than the alternative. And there is truth in that. Legal reasoning is genuinely complex. Precedent matters. The difference between how a statute is interpreted in one context versus another can hinge on a single word, and that word might carry a century of case law behind it.
But there is also a less comfortable truth, which is that the complexity serves the intermediaries. The language is convoluted not only because the law is nuanced, but because convolution creates dependency. You cannot represent yourself because the system was not built for you to represent yourself. The lawyers are intermediaries for arguments. The jury is an intermediary for facts. Every layer between a person and the resolution of their dispute is a layer that someone earns a living from.
Traditionally from first principles, the model was simple. Two parties bring their dispute to a king or the presiding magistrate. The magistrate hears both sides. The magistrate decides. This is how disputes were resolved for most of human history. The village elder, the tribal chief, the king’s court. The system worked not because it was sophisticated but because it was direct. The intermediary layers we’ve built since then were added for legitimate reasons, to handle scale, to reduce the power of any single decision-maker, to protect against bias. But they’ve accumulated to the point where the system now serves its own machinery more than it serves the people who enter it.
If AI makes the discovery of facts and the construction of arguments radically more accessible, then the necessity of many of these intermediary layers starts to collapse. Not the judges but every other intermediary including the lawyers. The human at the centre making the decision still matters, as a humanly check on the system.
None of this is technically impossible. The tools exist or will exist shortly. But the legal profession is one of the most entrenched intermediary classes in any society, and the system will not simplify itself out of existence voluntarily. The people inside it are not irrational. They are protecting what they’ve built and the prestige that comes with it. That is the problem.
The corporate version
The same logic applies to companies. And the same impossibility.
Right now, a business that needs a system has two options. Buy something off the shelf that does approximately what they need and spend the next year bending their operations to fit the software. Or pay a consultancy a small fortune to build something custom, wait months for delivery, and hope the result matches what they described.
AI changes this equation. Not because AI is magic, but because the speed of building changes what’s possible. If you can go into a company, spend a week understanding its actual operations, its actual pain points, its actual workflow, and then build a system tailored precisely to those needs in hours with Claude Code rather than months, you’ve eliminated the core inefficiency. The company doesn’t buy a platform designed for a thousand use cases. It gets exactly what it needs. No admin to configure it. No integration tool to bridge it. No analyst to interpret it. The software fits the operation, rather than the operation contorting to fit the software.
This is what AI building speed actually enables. Not another SaaS product for the pile. The opposite. The ability to treat software as something bespoke, crafted for the specific context, and evolved as the context changes. A company that can build and iterate its own systems internally, shaping them as the business shifts, is a company that doesn’t need to accumulate the parasite layers of administration, integration, and maintenance that the current model demands.
The promise is not more technology. It is less intermediation. But less intermediation means fewer intermediaries. And intermediaries vote, lobby, and resist.
The fractal problem
Nassim Taleb, drawing on Benoit Mandelbrot’s work, describes a model of the world where the same patterns repeat at every scale (I wrote about this previously). A fractal is a shape that looks the same whether you zoom in or zoom out. A coastline is jagged at the scale of a continent and jagged at the scale of a single rock. Mandelbrot showed that financial markets behave this way. Taleb extended the idea to fragility and disruption more broadly.
Apply this to organisations and the pattern holds. A small, nimble company overtakes a large, slow one because it is faster, hungrier, and unencumbered by the very systems the larger company built during its ascent. The incumbent is trapped by its own success, by the infrastructure that enabled its dominance, by the people whose livelihoods depend on the complexity continuing. Resistance to simplification is not irrational. For many people inside these systems, simplification is an existential threat.
This is what happened to Kodak, which invented the digital camera and then buried it because the existing business couldn’t survive the transition. It’s what happened to Blockbuster, which had the chance to buy Netflix and passed because the existing model was still profitable. The organism protects itself even when protection means death.
Scale this up and the same pattern applies to countries. To world orders. The civilisations that rise on lean systems and clear mandates eventually calcify into the bureaucratic organisms described above. The Roman Republic became the Roman Empire became a bloated administrative apparatus that couldn’t defend its own borders. The British Empire built the most sophisticated governance infrastructure the world had ever seen and then couldn’t adapt when the colonies no longer accepted the premise. The pattern is fractal. It looks the same at the scale of a company, a government, and a civilisation.
AI accelerates the cycle. It gives the nimble an even greater advantage over the entrenched. A two-person company with AI tools can now produce the output of a fifty-person company from five years ago. A country that embraces the change can leapfrog one that resists it. The gap between those who adapt and those who don’t is widening faster than at any point in modern history.
But you can see the future clearly and still not be able to get there fast enough. Society does not move at the speed of technology. It moves at the speed of generational turnover, of institutional inertia, of the deep, slow currents of culture and identity that determine what people are willing to accept. The tools are arriving faster than the readiness to use them. The frameworks are available before the mindset to adopt them.
I suspect that the societies and organisations that navigate it best will not be the ones with the best technology, but the ones with the least to protect. The ones light enough to move. The ones that haven’t yet built the parasite layer so thick that cutting it feels like cutting into bone.