AI Disruption and the Software Bloodbath

How to think about investing in software during the AI transition

The indiscriminate selling in software has been relentless, but it also presents an opportunity for investors who are willing to dive below the narrative of AI as a software business killer, into the reality of the impact.

Let’s look beyond the obvious counter-narrative points about enterprises moving more slowly, sunk cost in enterprise SaaS, retraining staff, and capital expenditure on AI infrastructure build-outs. One, because those horses have been beaten to death, and two because those things didn’t prevent companies from spending the time and money to refactor workloads for the cloud, or for cloud consulting services and development staff. For all their virtues, they have not been convincing enough points to cause investors to pause and think through the nuances of this particular situation.

The enterprise shift to the cloud taught us that when the incentives are great enough, enterprises are increasingly willing to spend resources on re-doing things.

Even though that’s the case, the transition from on-premises software to cloud-native taught us something else- there are always winners and losers. How do we tell the difference?

The most important way to answer that question is to consider what unique threats AI presents to enterprise software companies based on the users and use cases those companies serve.

Three-Layer Positioning
Let’s frame up what I consider to be the most fundamental shift with AI in its current iteration, which is its use as a new type of interface.

If we think of enterprise software in very simple terms, its purpose is to allow the storage of -and interaction with- a subset of an enterprise company’s information. We can imagine it has three layers.

1. The user interface is the surface layer.
   
   This is where we type our information into forms, where we search for information, and where information is displayed for us. It’s our interactive window into what the software is supposed to do for us.
   

2. The search, read, and write functions are the middle layer.
   
   This layer handles the exchange of instructions between what the user wants to do in the user interface layer, and the information stored in the deepest (database) layer.
   

3. The database is the deepest layer.
   

   It’s like a giant warehouse where all the information for a piece of software is stored on different shelves, and the place from which data is retrieved when a user requests it.

You may be wondering what this has to do with AI disruption, and here is the answer: AI has the potential to disrupt existing software at all three layers in the future, but the most immediate disruption potential exists at the surface -the user interface layer-right now.

Remember, timing is everything. If enterprise companies are busy replacing software that holds the most value in its surface layers first, it gives software vendors offering deeper layer solutions more time to adjust to AI adoption and to potentially leverage it to their advantage. Therein lies our opportunity as investors.

To illustrate what I mean by surface layer user interface replacement, take the example of working with an AI chatbot like ChatGPT or Claude. You type in a question, and it serves up an (hopefully accurate) answer in the time it takes to sip your coffee. It’s much more time effective than going out to google search and sifting through results of mixed relevance, especially as you ask the AI bot subsequent questions. It hasn’t necessarily replaced the deepest layer -the database layer- because the information is probably about the same as what you could find on your own if you wanted to work hard enough. But it’s greatly reduced the friction between you as the user and the information you’re trying to find. Even though this example is more consumer focused, for the vast majority of non-technical people making decisions about how to use AI within their organizations, this is a use case they’re thinking about.

NOTE: As I said before, AI has the potential to disrupt the other two layers as well. It can do many things in the background without human intervention. Cybersecurity has strong examples of this. I will write about that in a separate entry in the future.

The following are some questions I use to evaluate my stance on a software company’s exposure to AI disruption based on their users, use cases, and who they’re selling to.

1. Which software companies are most valued for their use as a user interface for working with a subset of a company’s data?

2. Who are the early adopters within an enterprise organization to leverage AI coding tools to improve their work experience?

3. Who is least likely?

4. Which existing software companies are best-positioned to accommodate new AI workloads, both technically, and from a business model/pricing standpoint?

5. What AI-driven activity might happen within a software vendor’s customer base for its earnings growth to either slow or expand?

I’ll use some real company examples to highlight my thinking.

NOTE: As we start talking about specific companies, I want to add a disclaimer here that this is not investment advice, and more importantly, I am not saying that companies I see as better-positioned than others will be able to actually capitalize on that opportunity, and similarly, I’m not saying that companies I view as having more exposure to AI disruption won’t figure out a way to spin it in their favor. But we have to work with what’s going on today, and I think this article would be pretty useless without some real examples of what I’m talking about. With that out of the way, let’s proceed.

Potential Winners - Database Companies and Deepest Layer Defense

ESTC, MDB

From my perspective, database companies like MongoDB MDB 0.00%↑ and Elastic ESTC 0.00%↑ are well positioned to benefit from new AI workloads. I worked in software for a long time in various technical roles, and there was a time in my life where I couldn’t be convinced there was a wide-spread, enterprise-scale use case where an unstructured database like MongoDB was superior, or even on par with a SQL database. But times have certainly changed, and the volume, structure, and processing of massive amounts of data has also changed. AI may finally be the nail the NoSQL database hammer was searching for all along. These database companies have also added relational data and SQL-like query capabilities to their platforms which have broadened the use cases they can effectively serve.

Widespread adoption of these databases, open source versions that allowed for early experimentation, and their scalability all set them up for success in the AI world. If a developer is working on a proof of concept for a new AI use case, nothing can slow down time-to-market faster than having to learn a new database technology. The wheel does not need to be re-invented in databases right now, the wheel everyone is focused on re-inventing is the user interface.

Perhaps equally as important is that these database companies don’t have to do much to revamp their business or pricing models to suit the influx of new data from AI workloads. Having to accommodate this change doesn’t slow them down in any way. They are already priced for data ingestion and expansion, and additional data works to their benefit. Retro-fitting a new pricing model during a technological sea change and expecting customers to get on board is no easy feat, as software vendors in other categories are likely to find out.

Potential Losers - User Interfaces for Technical Teams

TEAM, DDOG

Software vendors that offer what are essentially user interfaces for data or workflow within technical teams are in the most trouble. Particularly if they are very expensive or difficult to work with.

They are facing several headwinds.

Technical teams are the earliest experimenters and adopters of AI, and can easily understand its best use cases and limitations. It’s much easier for an enterprise company’s internal development team to build a replacement for some of Atlassian’s functionality, than it is for a non-technical sales and marketing team to build a replacement for the product managing their inbound leads.

If “software developers can rebuild software faster” doesn’t sound like a profound insight, I’d like to point out the obviousness of the statement is not reflected in software stock prices as they sit today. There is no differentiation happening as software names continue to get sold across the board.

Further, when a product is used by focused teams, rather than cross-organizationally, it’s easier to plan and measure a replacement. What will it cost, what will it save, how much easier will their lives be if they don’t have to use this product’s interface any longer? Much easier to sort when you don’t have to run across HR, Sales, IT, and every other team a workflow routes through to figure out the answers to those questions. The security complexity becomes more manageable as well (an entry for another day).

Time Will Tell - Cross-Organizational User Interfaces

CRM, NOW

Companies that sell to non-technical audiences, and/or are used more widely across the organization than just within technical teams are likely to fair better. Here is why.

Replacing an entire piece of enterprise software touching multiple departments across the organization is extremely disruptive to businesses processes and productivity. Not only that, but predicting cost and setting budget becomes much more challenging as well. You may be wondering why I don’t put these companies in firmly in the “Potential Winners” category. I believe it’s more likely small use cases will slowly get peeled off from these vendors and simplified by AI over time, which will feel like paper cuts to companies like Salesforce and impact their growth rate, but will not create the immediate disruption exposure cross-organizationally.

A bigger challenge to these companies is going to be hanging onto seats, and worrying about how they might leverage AI to make their products less cumbersome to configure and more user friendly over time before something new and shiny comes along solidly enough to cause a real problem. I believe they have time to do it.

Final thoughts

The question that emerges from this framework is:
At what price is it safe/smart to own a software company based on its three-layer positioning and current valuation? Blood is certainly running in the streets.