Bloomberg columnist Matt Levine sometimes says that markets love completeness - if there are things which are bundled together, markets want to find ways to trade them individually.

Employment is currently a big bundle of different properties, and some of what we have seen over the past few years is the unbundling of some of those features. A business is a series of activities coordinated to generate a profit, and the mix of how the necessary activities get done is, according to Coase and others, decided by the transaction costs of the different bundles. A business might hire a manager to look after a facility, contract with a vendor to have the facility cleaned, and hire a freelance graphic designer to put together the signage. Socially, we strongly privilege some of these over others.

Relations between companies and individuals are generally regarded as being in a few different forms: Owner/Shareholder, Employee, Non-employee work contractor, piecework and so on. Each of these is a general category for a range of options though. For example, is a member of the board of directors an employee? Is the zero-hour contract teenager at a fast food store in the same employment category as the troubleshooting consultant brought in to revamp a failing product launch? Internationally, we see other differences, such as between career track employees and non-regular ones in Japan. Over the last few years the categories have got murkier as the lowered transaction costs from digital technologies and smartphones added new options to the mix.

The bundle involves not just the work-to-be-done part of employment that the hiring companies most care about, but a bunch of policies around work - particularly how benefits like sick pay, healthcare, and retirement are funded and managed. One of the problems with approaches like California's AB-5 is that they view the world through a fairly binary set of employment relations, and so impact everything from ridehailing drivers to freelance journalists. Policy would be better served by finding ways to more cleanly separate the responsibilities, so that a range of employment types could contribute to the outcomes policymakers are looking for rather than them attempting to move specific relationships in to one or other arbitrary bucket.

We're still figuring our how marketplaces and matching work here: you might use an app to call a Lyft or to hire a window cleaner, but the relationship you have can look very different if you are (legally) contracting directly with the window cleaner. There are new venues and places for work too - fully remote companies like Gitlab are now not all that uncommon, and we are seeing work done in virtual spaces, mainly games, making the territory in which the employment takes place more ambiguous.

The other part of the lowering of transaction costs, though connected to the unbundling, is the intermediating of management by algorithm. Part of what made hiring people necessary was complexity of instructions: it's easier for a chef to describe to a line cook how they want a vegetable prepared than call up suppliers for that preparation. In many industries, the "manager" is now giving conditions to a machine, which translates those into individualized instructions for workers.

In many ways, that's what the ridehailing apps do when they give drivers routes and pick ups - no human is asking them to do that, but at the same time there is nothing inherently inhuman about receiving the instructions: a taxi dispatcher did roughly the same thing. The interesting question is around who gets to define the conditions, and how they can create and tailor those kind of logical processes. How can you, as someone with insight and experience, direct an "instruction generating" algorithm to distribute work intelligently to others, at a much greater scale than you could enable through human intermediaries.

In Chaos Monkeys, Antonio García Martínez suggests that every job will either be giving instructions to machines, or taking them from machines. The reality is likely to be messier: many more people will follow algorithmic instructions in one sphere, but also be responsible for generating them in another. The line of where an employment relationship begins and ends will only get fuzzier.