If you think the space of AI product development and funding has been a frenzy, you ain’t seen nothing yet.
Years ago (apparently 10 years ago back in 2015!?) I read a post by Jerry Neumann (“The Deployment Age”) that really resonated for understanding how major technology shifts evolve and the dynamics behind it. That post was about Internet and Communication Technology (ICT) being in the “Deployment age” – referencing the work of Carlota Perez. I’ve been thinking recently about that post and about how the framework could apply to AI.
I highly recommend reading Jerry’s post for a greater in-depth summary of the theory and its application to modern technology, which I’ll be quoting below but not trying to repeat.
At the highest level, the framework splits major technology cycles of “Techno-Economic Paradigms” (TEP) chronologically into two halves, the Installation phase and then the Deployment phase. The Installation phase is where the initial innovation, irrational exuberance, and bubbles happen and the technology looks for what you might think of as Product-Market-Fit writ large. The Deployment phase is where the technology is pulled through and widely deployed into industries.
For AI Technology we are clearly in the Installation phase, which itself split into four phases, with the first two as described here from the relevant Wikipedia page (Technological Revolutions and Financial Capital):
Irruption phase: There is an intense funding of innovation in new technologies. Clusters of new revolutionary inventions appear. New industries are established, and the construction of new infrastructure begins.
Frenzy phase: Increased speculation and financialization leads to a decoupling between financial capital and production capital. Capital is invested more in financial innovations than in technological innovations. Asset bubbles are inflated.
After hearing back from Perez I got to trying to consider the thought a little more fully.
And I agree, it is really complicated! And, I’ll add that “I’m no expert”.
But, I’ll try to argue briefly for two things here:
- There’s a next major technological revolution at play, a new TEP, not just another feature of ICT
- That the new TEP will soon enter the Perez “Frenzy” phase due to 2 aspects of deregulation
Age of Intelligent Automation
Perez theory so far covers five technological revolutions, the most recent being the “Age of Information and Telecommunications”.
I see that we’re in the midst of a sixth TEP we can identify, which I’d argue we call the “Age of Intelligent Automation”. It’s largely things that are being lumped under the heading of “AI” but it’s broader than chat bots and integration into existing digital products. I see it as encompassing the places where NVIDIA is putting their hardware – Digital Product AI (chat bots, etc), Robotics (complex generalized, eg. humanoid), and Transport (full self-driving).
(in 2017 IEEE coined and used the term Intelligent Automation, which I think is consistent with how I’m using it, but can’t guarantee the conceptions share exactly the same scope)
Why isn’t this just another development within the existing general ICT TEP? It is certainly built on top of the previous technology, which forms the basis for the new infrastructure. The features that Machine Learning in general enabled did fit within ICT even though they were innovative partly because they didn’t really create their own investable category. In contrast, the newer generation of AI capability moves the needle on functionality enough that it is an investable category of its own. More importantly there are new types of products enabled which were not possible before. And the potential impact on how businesses operate is larger, not just adding efficiency but changing the way work happens more fundamentally.
When did this new TEP start?
2012 is when the shift to “deep learning” took root, with AlexNet demonstrating unprecedented capabilities in image recognition.
As widely known, the Attention mechanism in deep learning (the “Attention is All You Need” paper) in 2017 developed within Google, is what kicked off the revolution in generative AI capabilities of Large Language Models, though it was 5 years later in 2022 that OpenAI’s ChatGPT really accelerated the availability of these new LLM capabilities broadly. And it’s hard to imagine that we’re only a few years after that.
Google’s self-driving car project started in 2009, achieving first self-driving on public roads in 2015. Tesla’s FSD really seriously started in the 2016 range or so. Today in 2025 people are getting automated cab rides with Waymo in some cities, and using FSD in their Tesla with increasing trust.
Given all that, the start of this TEP could be considered to be between 2012 and 2017, or between 8 to 13 years ago. I don’t think these things are exact, so we can say we’re about 10 years in at this point.
Why “Age of Intelligent Automation” and not “Age of Artificial Intelligence”? I think the primary for the age if we jump forward and imagine looking back in terms of impact is about automation – whether it is automating away the task of driving a car or of putting together a report. Whether we are automating acts (like driving) intelligently or automating intelligence itself (“reasoning” and report writing). Artificial Intelligence is more the means, whereas automating generalized capabilities is more the end.
Irruption
Reminder of the Wikipedia definition:
There is an intense funding of innovation in new technologies. Clusters of new revolutionary inventions appear. New industries are established, and the construction of new infrastructure begins.
I don’t think there’s much argument countering that this describes the last 10 years or so. We’ve had revolutionary inventions and construction of new infrastructure. One aspect of the infrastructure is the hardware itself, especially by NVIDIA, and the companies that have built AI clusters for training and inference with it, like Google, OpenAI, Facebook, AWS, Tesla. In this technology paradigm I’d argue there’s also a new type of infrastructure: the availability of open sourced models and platforms (eg. HuggingFace) for distributing them. You could argue there’s even a third type of infrastructure that has been developed, which is the availability of neural net capable devices (GPU/TPU/NPU) on end user devices becoming standard.
Another aspect of the irruption phase is capital being a mix of production and finance capital, which is definitely in play here. There has been a massive amount of production capital deployed by the likes of NVIDIA, Google, Microsoft, Facebook, Tesla, AWS, as well as financial capital from VCs into companies like OpenAI and Anthropic, plus VC into smaller companies creating new capabilities on top of the nascent infrastructure (Perplexity, Cursor, etc.).
Frenzy
I’m speculating that, only 10 years in, we’re at the edge of entering the real Frenzy phase now, especially in the Digital Product AI category, given the state of the technology combined with the marked change in regulatory landscape we are entering at the start of a second Trump presidency.
I like Jerry’s explanation of entering the Frenzy phase:
Seeing the money to be made, money starts to pile into the new technologies. As the scale of the new opportunity becomes more evident with each successful investment, investment in the sector grows exponentially. This is the ‘frenzy’ phase of the cycle. For example, the small venture capital industry of the 1970s lead to the much larger VC industry of the 1980s and then to the dot-com frenzy of the 1990s.
And the Wikipedia article points out the important role of capital in this phase:
Increased speculation and financialization leads to a decoupling between financial capital and production capital. Capital is invested more in financial innovations than in technological innovations. Asset bubbles are inflated.
There are two ways I think deregulation supports this decoupling of financial and production capital for AI, which kicks us into this Frenzy phase.
The first is deregulation in the overall startup space, opening up M&A opportunities, reducing risk in speculative investment, leading to increased funding for early stage companies.
Compounding the increase could be the effect of deregulation in the crypto space, enabling a whole new type of “financial innovation” that could be tied to funding of AI product opportunities. Some years back companies played with “creating a token” as a means of funding, which could come back with a vengeance.
A peek ahead at Deployment
I really wanted to call out this quote from Jerry’s 2015 post, because I think it illustrates well why general Internet and communication technology was and is in Deployment.
Stop considering the technology a feature. Using the technology where it fits is no longer a feature, it’s a requirement. Connecting a thermostat to the Internet wirelessly is awesome, but calling it an Internet-enabled thermostat will start to be like calling a vacuum cleaner an electricity-enabled broom. And if your thermostat does not connect to the Internet, it will be bought only by retro-chic hipsters.
We can both instinctively understand why that does apply to internet-connectivity and why it doesn’t currently apply to “AI”. But if we squint we can maybe imagine what it will be like to be in that far future that we are building where talk of “AI” has become passé.
Who Cares? (ie take-aways)
Understanding the cycle and where you are is useful when considering where to put your money, resources, and time.
Investment-wise, it’s not yet a bubble because we’re not actually in the Frenzy yet. This is the time to be investing more. Whether it’s money or whether it’s time.
For current or aspiring Software Engineers – there’s lots of talk about the role becoming obsolete, but the magnitude of the opportunities ahead have plenty of room for everyone to create innovative technology. Forge ahead. Double down (but evolve to use the latest tools).
Entrepreneurs, keep an eye out for innovative investment mechanisms that can maximize the value you get out of what you create. Also, consider that you are not selling artificial intelligence (the means), but rather you are selling intelligent automation (the ends).
Richard Shea's Blog 