How long will it take for AI to trickle through the economy?

I’ve recently been implementing AI in services businesses, which has got me thinking about how long it will take for AI to truly permeate through the economy.

I cannot emphasise enough for the SF tech people out there: the rest of the world is so far behind. In working with financial services businesses I’ve spoken to a few large software providers in the space whose software we wanted to integrate our system with. One of these businesses I am told spends $100 million a year on their tech budget - and yet does not have an API for their system, nor can they build one in less than a year (but it is on their roadmap to do, so I don’t think they’re trying to block me - giving data access is a core part of their strategy, they just move incredibly slowly). Another has had their entire very large tech team trying for months to build the agent setup that I built in a few days, and somehow failing.

It is clear to me that there is still a huge amount of alpha to be had in implementing AI into these types of businesses, and I suspect there will be for years to come, even when the technology has far advanced the point of capability to help them. I think part of this is job protection on the tech side - most of these types of organisations work with externally contracted tech teams, if they work with a tech team at all. External contractors generally bill hourly or by project; they have zero incentive to implement systems that would allow them to move quickly. Even internal tech teams in large companies have little incentive to explain to their non-technical higher-ups how much faster it is possible for them to do their work because of AI. (The people working in services businesses have much lower job risk, because in services businesses there is always more relationship work to be done. But software developers who just crank out code without thinking about systems and architecture as a whole are becoming less necessary quickly, and the response of a lot of software developers is to resist the change rather than learning how to become high leverage with it.)

On a side note, I think there’s a piece of this in a different form happening in some tech companies in SF - there are a lot of complicated-sounding technical terms being thrown around to describe a basic set of concepts that are actually pretty obvious and simple to implement. Tech companies there are also grappling with their own place in the world as AI improves, and the very human solution is to make what you’re doing sound very complicated and smart rather than talking about it in its simplest form. Then the non-technical people still think you have some magic. (P.S. Please don’t cancel me SF, I know there are lots of brilliant genuinely hard things being built in SF tech companies, and I’m sure some of the things I think are simple actually have their challenges to implement. This is just my current take and I may change my view at any point.)

I see three obvious ways for AI to trickle through the economy:

1. Existing people within these organisations, particularly younger employees keen to prove themselves to management, start implementing and talking about AI.
2. We see a lot of consolidation into businesses that are AI forward and therefore move faster than any of their competition.
3. Whole categories of businesses dissolve and are entirely replaced by people’s personal AI.

I suspect a mix of these things will occur, but they will occur over a much longer time scale than the actual technology development. There are services businesses still using physical servers rather than cloud-based software. When you own a lifestyle business that’s still generating returns, has good customer retention, and you’re coming up to retirement like many of these services business owners are, there’s very little incentive to go to the effort of changing the way your whole business operates to implement this new tech wave. In fact, you’re probably disincentivised because your team won’t like it and whoever your IT provider has been for 20 years makes your life a nightmare anytime you try to change anything technical.

Option 1 is starting to seem less likely based on young people’s current attitude toward AI (see the booing at college graduation stories). Option 3 may happen - but there have been a lot of technology waves that have made it possible for people to perform some of these services for themselves, and these businesses still exist. People pay for the privilege of not having to think about things, and it will take a long time before most people would trust AI for important services enough to not second-guess its opinion. To me, option 2 seems the most likely to happen quickly; particularly if opensource models start being widely used, a company with a high-leverage, well-incentivised internal AI team will still have alpha over businesses using just Claude or ChatGPT.

My frustration as a contractor is watching sleepy businesses knowing that option 2 is both an opportunity and a threat to them. There are some business owners who get it immediately, and when I work with them I quickly drive outsized value. But there are so many out there who never overcome their inertia enough to even look for this type of solution, and they don’t have the ability to identify or attract the necessary talent themselves. Without option 2, it will be a slow, slow process before AI provides all the productivity gains it can to the economy.