Digital Gig Workers: Exploring the Evolution of Software Pricing

‍

The nature of work is evolving. In human labor markets, we're seeing a shift towards the "gig economy" – or short-term, platform-mediated work by companies like Uber, Doordash, Airbnb, and Upwork. Between 2018 and 2023, the gross volume of this market more than doubled, now approaching half a trillion dollars. The trajectory seems clear: traditional employment is giving way to flexible, uncertain work arrangements.

Gig work often represents fairly commoditized functions - almost anyone with a drivers license and a car can become an Uber driver, or almost anyone with a house and a spare room can be an Airbnb host.  If you choose to exit the market, it would be fairly easy to replace you with another driver or homeowner.

This trend in human labor markets may give us a glimpse of how software pricing or “digital labor” markets will evolve. Historically, the difficulty of producing your own applications or hiring an ML team to develop specialized inference functions was steep, so traditional SaaS providers were able to charge high markups with recurring fees for access to their functionality. That’s changing quickly, as several companies are now producing Large Language Models (LLMs) that are approaching or exceeding the capabilities of knowledge workers.

Stable, recurring payments for software

Flat subscription based pricing emerged as an improvement to traditional software licensing models that required large upfront costs for perpetual licenses. With a subscription based model you could spread the costs out and lower the barriers to entry for consumers and businesses. That model sounded great in the 2000s.

With the proliferation of cloud computing scaling no longer required large upfront investments and was competitively accessible from providers like Azure, Amazon’s AWS, and Google Cloud. Scaling became a commodity that was priced on a continuous scale and that dynamic pricing trickled to consumers of those products. As a result we saw the increased popularity of tiered and seat-based subscription models in the 2010s.

While scale became a commodity, SaaS functionality built on top of this cloud infrastructure was not. Applications involving machine learning still required large upfront investments (you had to hire a team of ML engineers).

Pricing commoditized digital labor

The availability and rapidly increasing capabilities of LLMs has dramatically lowered the barriers to creating intelligent, inference-capable applications. What once required a team of specialized engineers can now be replicated by structuring interactions with a base model. These applications can not only easily replicate legacy applications but can increasingly act autonomously.

Their increasing resemblance to human labor, in particular knowledge workers, makes some speculate that these autonomous applications will be priced at a discount to their human counterparts. This model charges based on the time or units of work completed.

While agents may resemble human labor in their output, they are incomparable in their availability and scarcity. The price of human labor reflects how finite it is, so it is unlikely that historical labor pricing will be an accurate benchmark for agentic pricing.

With the availability of open source LLMs, the underlying models are increasingly commodities. With any commoditized market, the price trends towards the marginal cost of production - giving support to the cost-based model.

The new value benchmark is the underlying models themselves - and in particular the performance of the open source ones. If a base model provider like Openai tries to charge for a model that’s comparable in function to Meta’s Llama or another open source model then users could always switch to the open source one and get similar outcomes without the fees. Any premium in model pricing has to be for the increased performance above the open source counterpart.

In such a world, applications built on top would charge according to the value of their structured interactions with the underlying LLM. Since LLM interactions are priced according to input-output tokens, we’d see the emergence of token-premium models.

We actually see this with web3-apps where the underlying smart contracts are openly accessible and forkable, i.e. commodities, but organizations like Uniswap charge a small take rate for using their frontend to interact with the underlying contracts.

This token premium model would also resemble the general trend in human labor markets towards short, project based work. A distinction I see is that the human gig economy is skewed towards catering for consumers who need takeaway, a ride to the theater, or a place to stay while visiting New York. For businesses that have recurring needs, they may value cost certainty over potential savings and pay a premium for that stability. It will be interesting to see how it plays out.

‍