The AI Follies
(or, everything you always wanted to know about software but were afraid to ask)
“Curtain! Fast music! Light! Ready for the last finale! Great! The show looks good!” – Florenz Ziegfeld Jr.
“The fool: Before you know it, the Renaissance will be here and we’ll all be painting.” – from the movie “Everything you always wanted to know about Sex (but were afraid to ask)”
I remember when followers of several lines of code firmly believed their subject of adoration was going to revolutionize payments, currencies, means of exchange, store of value… Each state of rapture came with vociferous claims soon to be replaced by another rapture once previous claims were debunked by facts. There was nothing bizarre about such cycles as human beings need to believe, none more so than investors talking their own books.
I have recently observed a new rapture with its own vociferous claims, vocalized by a different set of followers who equally firmly believe in different lines of code.
AI peddlers will have us believe that new advances in AI have or are about to destroy our world more swiftly than it took for me to write this sentence.
Witness how publicly traded B2B SaaS companies have performed in the past twelve months – the SaaSpocalypse wiped out nearly $300 billion in market capitalization. Witness how, more recently, an announcement from Anthropic on its new Claude AI security tool wiped out over $15 billion in cybersecurity stocks.
The markets are clearly extremely sensitive to AI developments and are marking down the valuation multiples of many software companies as a direct result of more uncertain growth rates and market sizes going forward.
These market gyrations are nothing new and should be expected with every new piece of news being added to the aggregate knowledge already priced into stocks.
What puzzles me is how overreactive the markets are to AI news and how AI followers and their pundits are making grand statements. Some pundits are comfortable predicting massive deflation and demand destruction, see here. Although this piece is a scenario and not a prediction, it certainly caught the attention of markets and the AI stratosphere. The general argument – extremely simplified – is that AI tools will usher abundant intelligence which will in turn displace or completely replace existing software tools as well as humans powering and using these software tools (hence the SaaSpocalypse).
Although I fully buy into the transformative powers of AI innovation across most segments of our economies, I admit having a hard time with arguments that point towards a wholesale replacement of the workforce. Marx fell into the same trap when not taking into account how every innovation or capital investment carries within itself the option of creation and growth. His “capitalist hung with the rope he sold us” trope could never comprehend that each innovation wave manifests itself as an S-curve of growth buttressed by new employment creation.
To be more precise, it seems that we are stuck between a) those who claim AI will not live up to its expectations and its invested CAPEX to date, leading to a meltdown in the markets, b) those that claim AI is far better than anyone realizes and therefore will destroy every job (in the software industry and in all other industries) and c) those that claim AI is far better than anyone realizes and will lead to the end of scarcity and a golden age of abundance for all. No wonder markets are dancing to the AI follies. Who needs enemies with such a collective bunch of rabid pundits?
As always, the truth will end up being more measured. Every technology S-curve is confronted with its own limits and friction (we will not all lose our jobs immediately; we will not live in a world of abundance the day after tomorrow). Every technology S-curve ushers a supply dislocation with some job losses and invariably ushers in growth and job creation in ways we could not fully comprehend at the beginning of the process.
Be that as it may, it behooves us to analyze how AI will impact software. Indeed, the extreme counterargument which states AI will have little to no impact, is as erroneous as the maximalist one.
I came across this piece by Nicolas Bustamante and liked it so much I adopted its framework for our fintech purposes.
The framework is helpful in as much as it identifies 10 economic/strategic moats for any software (emphasis being given to vertical B2B software). The first five moats, I believe, will be weakened – the severity of which still remains to be clarified. The last five moats are still standing – some more so than in the past.
- Learned Interfaces: the user interface of a software is the product of years of experience coded for the benefit of the user. It was a moat where users would feel comfortable and where a user journey was scripted in its most minute details. This moat will be weakened, either materially so or severely so as AI tools allow anyone with subject matter experience to build UI blazingly fast and at a fraction of the cost.
- Custom Workflows & Business Logic: what sits behind the user interface will also be weakened for the same reasons the user interface is weakened.
- Public Data Access: Any AI agent can sift through, organize, analyze, and present public data much better than in the past. This moat for traditional software products will not necessarily be weakened further, although it will be completely commoditized.
- Talent Scarcity: Building software required subject matter experts and top engineers. Sometimes these people would be one and the same, hence the scarcity. Now, with AI code protocols, anyone with subject matter expertise and without the necessary depth of talent on the engineering side will be able to develop software. One more moat being eroded and, in this instance, inverted.
- Bundling: software products would expand over time, via new features, new functionality, new product bundling, new geographies and a mix of all the above. Now that many more people can code, this moat will also be weakened, although less so than the first four.
- Private & Proprietary Data: Probably strengthened given that data quality and scarcity make it even more valuable for AI technologies.
- Regulatory & Compliance: You cannot code regulation and compliance. Any startup building software or services around technology that needs to be licensed, supervised, regulated will continue to do so and be insulated by AI disruption.
- Network Effects: This is a structural moat that will remain potent going forward.
- Transaction Embedding: the closest one’s service or software sits next to a transaction, monetary or non-monetary, the higher the cost of replacing such service or software. Once one is embedded in or around a transaction and remains relevant, one has a bright future ahead.
- System of Record: If you build a system where the state of an asset, of a transaction, of a data point resides and shows the “truth”, being disrupted by AI is much more difficult. Core banking systems, core insurance systems, core payment systems, core systems for hospitals or health care in general. These tend to be very sticky and immune to rapid disruption.
Interestingly enough, the last five moats standing are all found in totality or in combination with fintech startups. This leads me to state that fintech may not be as vulnerable as other segments.
There are several implications, should the above become true.
First, market sizes will change in quantum and shape as many new entrants will compete in existing or new categories. This means multiples for valuing software startups will be under pressure, and the discount rate applied to early-stage startups will increase thereby concentrating even further investments on exceptional startups.
Second, existing categories in the fintech space will remain relatively immune (capital markets infrastructure, asset management infrastructure, stablecoins, tokenization, digital assets services, embedded finance…) for reasons explained above.
Third, new categories may emerge such as AI-deep startups with some regulatory license or heavy compliance attached.
Fourth, we should expect some short-term dislocation with startups exposed to AI innovation, where value propositions will resonate less with buyers. This in turn will have a negative impact on VCs who have traditionally invested in SaaS businesses. What is happening in the publicly traded markets with SaaS assets is only a precursor of what will happen in private markets.
Fifth, the way VCs conduct due diligence will change as the way startups build their technology and value proposition going forward.
Sixth, expect major disruption in how value will accrue going forward in the software space. Indeed, as AI agents will proliferate, we should not be surprised that machine to machine interactions will dominate software usage at some point. The subtlety of this statement lies in the fact that a machine does not “care” about how pretty or logical the user interface of a software is. Only humans do. A machine will only “care” about how precise the API of a software is. Therefore, it follows that most software which cannot remain immune or cannot be differentiated, will end up being abstracted as mere APIs and buried below AI agents and AI platforms. Implications for build, CAPEX, OPEX, pricing of said software come to mind. This does not mean the death of the user interface in the software world. What it does mean is that new user interfaces, residing in a different place and controlled by different actors will rise. The bottom line is, following aggregation theory, eyeballs/user activity/user actions will take place somewhere else, and be monetized differently than in the past.
Seventh, history has taught us a few things that apply to our AI discussion. Once the cost of something plummets, demand increases (sometimes demand explodes). The advent of Excel did not doom number crunchers, analysts and accountants – on the contrary it increased the demand for number crunching, analysis and accounting. So, it will go with AI. With the cost of building technology decreasing, I expect many more software products on the horizon (and many more jobs associated with these software products). We are all coders now; we are all AI experts now. We are also physical beings that like to interact socially with our fellow humans. The advent of the internet and such things as Zillow or digital insurance brokers have not killed real estate agents in the first instance or insurance brokers in the second. Some jobs will prove far more resilient than current AI maximalists believe. As it applies to fintech employment I expect the same resilience for certain categories.
My purpose here is not to convey that software as we know it is dead. Far from it. New models will take quite a long time to emerge, and we should not discount the competitive moves of software incumbents. Rather, it is to flesh out how and where value may accrue in the future and how investors may react to future batches of new startups.
The end game is somewhat clear: aggregation and value creation will follow different paths than during the traditional software/cloud computing/API age, as AI agents and AI tools are used broadly across our economies. What is much less clear is the timing, intensity and shape of said end game.
