AI Won't Fire You - It'll Gigify You

Everyone's worried about the wrong thing. AI isn't replacing workers - it's turning them into Uber drivers. Even lawyers. By 2030, you'll be bidding for 6-minute tasks on an app.

Hi there,

You’ll write your resume in YAML and expose an API endpoint on your server to find work in a few years. I wasn’t joking.

The end of the firm as we know it presents numerous interesting avenues for research. Examining how work is allocated today yields some valuable insights.

I've spent the past six months extensively experimenting with AI tools and conducting in-depth research, drawing on my background in financial services technology, change management, and my degree in economics.

Human coordination is expensive and complicated. AI leads to less human coordination of tasks and more machine coordination of tasks. The AI, API, and AI agent ecosystem can lead to tremendous efficiency, but there are very real social costs to this mechanistic future.

This shift is all about optimising a business so it is easy for AI agents to operate inside that business and optimise outside of it. This eventually means API endpoints for everything.

While many companies are claiming to invest in AI, few are mature in their deployment of the technology. Workers believe AI will replace a lot of their jobs, and they’re not wrong.

New jobs will be created, but it’s increasingly clear that the total number of jobs will decrease over time as these new realities spread to more industry sectors.

The skill level required to justify employment in the AI era could continue to rise, leaving more people unable to compete with machines and struggling to navigate life with no career ladder in sight.

A McKinsey report claimed AI would add $4.4 trillion to the economy. That’s not going to your back pocket.

Today, we’re looking at humans as API endpoints - you’ll need a Swagger file.

If that sounds like science fiction, gig workers already work long hours under algorithmic control - they're the ones who just dropped off your dinner.

The Precedent

Amazon used to have the “Mechanical Turk”. They called it “artificial artificial intelligence”. You could pay small amounts to have micro-tasks completed. In China, concepts such as “talent cloud” and “human as a service” have emerged on some platforms.

Algorithmic gig work already exists, and the profit incentives that led to the success of Uber and DoorDash will inevitably extend to all remaining human tasks that need to be completed.

The value that human workers add extends beyond merely completing tasks. Creative problem-solving, ethical judgment, humanity, and things that can’t be automated to human levels of satisfaction are all apparent gaps in the assumption that everything can be automated.

Some academic frameworks talk of people’s role in cyber-physical ecosystems as modularised work, performing tasks that machines either cannot or where humans retain a comparative advantage, i.e. it is cheaper and easier to get the human to do the thing than add a feature to an industrial robot, humanoid robot, or AI agent.

This “service-oriented architecture” view would be familiar to the technologists among you. Each task and capability becomes modularised, with clear definitions of what can and cannot be done inside constraints programmed in.

When it comes to programming how humans will respond to tasks, there is a lot of complexity that potential AI overlords may find frustrating - mood, fatigue, expertise, gut instinct, the need to take a rest break or annual leave. Yet, if they want humans to perform the residual task set, some mutual accommodations will need to be agreed upon.

The reality of the gig economy is both terrifying and sad. Algorithms assign, supervise, evaluate, and terminate. This reality of labour market participation is already here for the less fortunate. Knowledge workers panic about losing their jobs due to AI, yet they have largely ignored the working conditions of Uber drivers for a decade.

The stories of warehouse workers facing very aggressive “pick rate” KPIs or dark store retail workers picking online orders give pause. The remaining white-collar work, which used to be performed in nice offices with city views and well-stocked kitchens, is now allocated via an app.

Lawyers gathering around a courthouse on their scooters, waiting for their allocated 6 minutes of “courtroom task” if their ratings yesterday were acceptable. Consider your industry and the residual tasks that will be difficult to automate. Few people will be needed to perform these tasks, and hardly any will be required to assist the machines.

The transformation of work into a gig-like nature has consequences, including longer hours, increased uncertainty, and large businesses exerting significant power over both workers and consumers. The protection of unions and guilds could become one of the most significant political issues of the 21st century.

“I’m sorry, only a human can complete that task - it’s illegal for an AI agent to do that”.

The Transformation

The flat organisation has long been a consultant's dream. Few firms achieve this lack of hierarchy. The rise of AI capabilities means that supervision becomes less of an issue - a manager can have a wider span of control and become a “decision enabler,” focusing on empowerment and innovation instead of micromanagement at the task level.

The traditional firm exists because of the desire to minimise transaction costs and perform some tasks internally, while contracting others via the marketplace. AI drastically reduces these costs. When these costs plummet, unless there are proprietary data sets or platform and network effects, the justification for large firms starts to break down in some industries.

This may lead to firms focusing less on human customers and more on making it easy for AI agents to do business with them via API endpoints. Platform-based models will become the dominant power players in this ecosystem, with niche, value-added solutions built around their core capabilities.

Hyper-competitive creative destruction makes it challenging to predict exactly how this will unfold over time. However, it appears that firms will become less pyramid-like and even liquid organisations over time.

But what happens to humans in this new architecture? This is a two-tier system: the designers and operators of APIs and those who become them.

The Human Cost

There are emerging jobs in the new division of labour. I’m not quite sure what an AI whisperer might be, but there are training, explaining, and sustaining types of roles that will emerge.

In technology, even if AI coding agents become dominant, many human tasks will still be required for the time being. Taste and human interaction come to mind. As industries like manufacturing become increasingly automated, the skill bar required will continue to rise.

The transformation of existing work will look different across domains. In finance, AI will handle both routine and complex tasks, and regulatory interactions will require human attendance at regulatory meetings.

In software engineering, tasks such as testing, architecture, and design will continue, alongside complex debugging and problem-solving. Over time, more and more of these tasks will be automated, but as costs fall, the insatiable demand for more software makes this a challenging category to predict where things will ultimately end up.

The role of management and the C-suite will evolve. Less supervision and micromanaging, more strategy and coaching. One aspect of this shift is that technical executives will become dominant as leveraging AI technology becomes the general-purpose commercial dominance strategy.

I’ve mentioned before, and many commentators also mention, that empathetic and ethical judgment remain core human tasks. Physical presence in highly consequential decision-making loops, especially in military or policing applications of AI, will still be necessary. Profound levels of ambiguity or deep philosophical questions could occupy much of the future labour force’s time.

The Power Struggle

Estimates of the economic value or productivity improvements that AI will bring are in the trillions of dollars. Whether these eventuate is another story, but the benefits will be unevenly shared. In a world where top AI researchers can get $100 million signing bonuses to join Meta, there are also gig workers competing for the next food delivery task.

The challenge with the rising capability of AI models is that, task by task, function by function, capability by capability, and company by company, fewer and fewer human workers are required to complete the residual tasks that can’t be automated yet.

The real power dynamic hinges on who controls the API endpoints. If you control your own, and there are regulations on human task provision similar to employment rights today, that is one scenario. That is highly implausible, though, because that’s not the status quo today!

If AI agents become autonomous in their loss minimising, many will have no issue employing humans to complete tasks in a race to the bottom. If more B2B activity shifts from human-to-human to machine-to-machine, the rights we have taken for granted as a society for 100 years may come to a swift end.

Regulation won't save us. Regulatory capture and lobbying will ensure AI serves Big AI and capital, not workers.

The Default Path We’re On

Which future are we heading toward? Based on current trends and incentive structures, I have identified four scenarios, along with my corresponding probability estimates.

Consider two key issues: who controls the AI (the power and tech are either centralised or decentralised) and how many humans are truly needed (labour is scarce or abundant). These four scenarios form a 2x2 matrix based on who controls the AI and the number of humans required for operation. I assign my own estimated probabilities of each scenario unfolding by 2030.