AI is yet another platform shift

The debate over whether AI will displace job loss predates the latest AI hype cycle, and frankly, it’s not a particularly interesting one in our opinion. We view AI as yet another platform shift in the long history of technological evolution, going back at least as far as the printing press. As with every previous platform shift, AI will automate some portion of the work that humans currently do, but far more importantly, it unlocks a massive amount of latent demand.

This is Jevon’s paradox. In economics terms, as the cost of producing a good drops, the corresponding drop in price causes a surge in demand. If we had to summarize this post in one sentence, we’d say: Jevon’s paradox has followed every platform shift in history, and AI is yet another platform shift in which the surge in demand will occur.

Here’s how ChatGPT defines a platform shift:

A platform shift is a major change in the underlying technology or system that a company or industry relies on, often driven by technological advancements or strategic decisions. This can involve moving from one type of software, hardware, or operating system to another. Such shifts typically require significant adjustments in operations, development, and strategy.

To illustrate what we mean, let’s forget about AI for a second and go back to the printing press. Prior to the printing press, scribes were considered highly-skilled workers who had to be both literate and artistically-capable. The advent of the printing press, for example, shifted demand from scribes tediously hand copying the same document repeatedly to new professions as the cost of distributing information plummeted. Prior to the printing press, people wanted information, but spreading information was so cost prohibitive that it was unavailable. When the cost of production dropped, writing and literacy became realistic, and new opportunities emerged for what we think of today as authors, journalists, and teachers.

An example that might be more familiar is the impact that the internet had on communication. The creation of email and later of instant messaging dropped both the time and the dollar cost of communication. Off the top of our heads, internet-based communication cut demand for USPS, printers, and maybe handwriting skills. On the other hand, it unlocked a massive amount of latent demand for low-latency, low-friction communication. Cheap communication clearly changed social relationships, and it also meant that businesses could get the word out for a fraction of the cost. Over time, a new kind of sales and social media marketing skill emerged — something we could have barely imagined 30 years ago.

Ultimately, AI is yet another large platform shift. We’re not trying to downplay the importance of AI — it’s already having a massive impact on our lives — but our expectation is that the evolution of the technology will follow a similar pattern. Dramatically reduced costs for work will create a surge in latent demand that didn’t exist before, which will increase consumption while unlocking new areas of focus for people.

It’s important to establish exactly how AI unlocks latent demand. The short answer is that it looks a lot like what we described above: The cost of doing certain tasks that were previously done by humans (who were compensated accordingly) has dropped by orders of magnitude.

Take RunLLM as an example: By offloading the tedious process of answering repetitive customer questions, support engineering time is unlocked to build customer relationships and provide architectural guidance that ensures customers are successful. More importantly, what we’ve seen with RunLLM is that it unlocks an incredible amount of previously-unserviced demand. Before RunLLM, users would only ask questions when they were absolutely stumped or when they believed they wouldn’t be judged by the other person. With RunLLM, you can ask a question without fear of judgment and get an answer instantly. The cost has come down by orders of magnitude, and the demand has correspondingly skyrocketed. We’ve seen communities like DataHub or Livekit go from answering a few hundred questions a month to thousands of questions a month with RunLLM.

Another illustrative example is Devin, which we wrote about last week. The technology might be too early to fully deliver on the promise of an AI-powered software engineer, but it’s clear that dramatically reducing the cost of writing code means that more products will be built and with higher quality.

We don’t have the foresight to predict exactly what new opportunities these changes will unlock, but it’s pretty clear that most of AI products can automate tedious work and unlock latent demand without replacing humans altogether.

Assuming we’re right about AI being a platform shift, why should you care?

1. Get on the train, or get left behind. If you’re reading our newsletter, you probably agree that AI is important, but platform shifts have a tendency to create a new crop of winners and losers. The shift is already happening, so we should all be asking ourselves how we can better leverage AI. It’s worth noting that this absolutely means there could be disruptions in the transition period, due to the failure of businesses that don’t adapt — however, we believe that (as with previous platform shifts) the demand will shift into other areas that are less tedious for people to manage.
2. It changes how products are built. We’ve discussed this in the past; the quick summary is that AI products should be built to be accelerants — not replacements — for humans. As you’re defining your workflows and thinking about your integration points, it’ll behoove you to think about how to build human-in-the-loop workflows rather than fully autonomous ones. When you start thinking about efficiently doing AI-to-human handoffs, you have to ask yourself questions like “How can the person get up to speed ASAP?” and “How can my system learn from what the person did?”
3. Integration is everything. While platform shifts always create new winners, there’s a clear necessity to integrate with existing products. Smartphones took off when custom apps were possible, and those apps in turn enable the explosion of products like Facebook and Instagram. Cloud infrastructure grew by enabling users to use familiar systems (Linux, Postgres, etc.) for a fraction of the work. The same is going to be true here — the most successful AI products will pull in data from everywhere your business already works.
4. Trust is everything (and so are vibes). Throughout 2024, we focused on measuring how much better RunLLM performed against other solutions. But what we’ve realized is that nothing matters more than building trust with customers—a challenge that’s harder to quantify but comes down to prioritizing quality and taking feedback seriously. For demand to keep pace with the growth we’re already seeing, every AI product will need to make this a relentless priority. Never compromise on your customers’ trust.

The promise of AI isn’t waving a magic wand that’s going to make everyone’s lives dramatically different. Just as with previous platform shifts, the promise is getting boring, tedious work out of the way to let people focus on the things that are the most exciting to them. That’s something we should all be excited about, and it’s something that we should all keep in mind as we’re building new AI products.

Thanks to Peter Farago for input on the framing of this post.