AI vs. The Steam Engine: Fight Me

AI vs. Steam Engine

If you are looking for a doomsday treatise on AI’s impact on the software industry (or the world for that matter), you’re in the wrong breakout room! Instead, I thought I’d share a framework for how we think about the emergence of ever more impressive AI and how we at Underscore VC think about its trajectory and its impact on our industry and the world. First, some spicy clickbait:

“The End of SaaS is Closer Than You Think” – (SaaS is already moving towards usage-based pricing…just like AI.)

“AI is Going to Eat Software” – (AI can’t eat…yet. And, oh yeah…AI is actually software too.)

Doomsday Clock’ at 90 seconds to midnight amid nuclear and AI threats – (AI is at 3:45am.)

Can a Fluffy robot really replace a cat or dog? My weird, emotional week with an AI pet” (OK, this is awesome.)

But somehow, this supercycle-driven hype manages to both overstate and underestimate AI’s real impact on…well…everything. Here’s an oldie, but a goodie:

 

So, with that as an amuse bouche, let’s look at how we think about where we are now and what the next several years of AI shoe-dropping will bring.

Baby steps to quantum leaps

When LLMs and Gen AI kicked in the door a few years ago, it was immediately clear that, from this point forward, ALL software companies are now AI companies. This isn’t just the musings of an obnoxious VC asking a founder about their AI strategy, but rather a reflection that AI must be — now and forever —  a part of every software tech stack. The capabilities and tooling are simply too powerful, ubiquitous, and inescapable to ignore.

As hard as it is to capture this upheaval in a single slide, this one helped us crystalize how we think about AI investing during this supercycle. We started by organizing the AI capabilities into “work to do” buckets. Despite the “waves” metaphor, the x-axis here is not really meant as a timeline — especially since much of this is happening simultaneously! That said, it does roughly reflect “share of work” and the impact AI will have on people and businesses in the coming years.

 

In Wave 1, we have AI capabilities that are really features of a larger solution. For example, being able to talk or type instead of pushing a mouse around means AI is part of an app’s user interface. Not only is this often much more efficient, it’s a big step forward in accessibility for many users. AI assistants and co-pilots are significantly more interesting, but are mostly still a feature of an existing app like GitHub’s Co-pilot or Gmail’s auto-complete tools. Depending on the use case, there can be huge productivity gains, but AI is secondary to the app’s primary mission.

Wave 2 is where much of the founder innovation and VC investing heat exists right now. We call this wave “native”, or “AI-first” because AI is more than just a feature — it’s the whole point of the platform. For example, AI is awesome at understanding complex, repetitive processes, calls to external systems, analyzing data, and using NLP and computer vision to make sense of messy workflows. These workflows exist in every business and industry and are ripe for AI automation. And unlike the rules engines we’ve used historically for some of this stuff, AI can learn and improve as it works. The takeaway? Every existing SaaS platform will need to adapt or die (many will do the latter) and every new platform will need to lead with AI.

The ability to ingest vast amounts of data, leverage inputs from multiple systems, identify patterns and trends, and ultimately make recommendations means AI is quickly crossing the “decision-making” rubicon. The number of system inputs and the amount of data an AI can parse and “understand” vastly exceeds human capacity. This means that the insights, predictions, and recommendations generated by a well-constructed AI can, at a minimum, dramatically augment any human-led decision-making process. As these AIs become increasingly autonomous, the promise of “agentic AIs” that act on behalf of their human handlers will be fully realized.

Wave 3 is the simply natural outcome of ever-increasing AI competence. It is also the source of well-deserved hand-wringing from economists, sociologists, ethicists, and philosophers. We are at the very beginning of a historical technological evolution that will leave little untouched. To be more specific, it is certain that jobs, companies, and even entire industries will become increasingly automated. It is also certain that, as in times past, this historic evolution will bring efficiency, productivity, and new jobs along for the ride.

We are seeing this change in real time as firms decrease or pause hiring while they lean into the opportunity AI represents. Every software engineer we know uses AI to help with coding tasks — and this “help” represents more and more of their daily work. It is also easy to imagine entire AI businesses — outsourced AI call centers, for example —  that rely more on AI than humans. In fact, entire industries will likely be made (or remade) from an AI-first perspective. Some sectors of financial services, like insurance, rely heavily on algorithmic decision making and customer support reps. Could an AI do most of that? You betcha.

When everything is AI, where does the real value get created?

In a universe where we all have access to the same foundational models and AI toolsets, founders need to think carefully about where the real value will accrue for their business. Our high-level rubric for investing in the age of ubiquitous AI is pretty simple:

 

Data is the gold, the oil, and the spice of the AI Age. And AI’s appetite for training data is insatiable. We are starting to see the plateauing of foundational LLM advancements as restrictions on data access across the web take hold. The good news is that businesses generate huge amounts of their own data (way more than the publicly available data sets) that can be used to adapt foundational models to new use cases. At Underscore VC, we like startups building for customers with huge data sets that no one else has access to.

Agentic AIs are particularly valuable where there are hard business problems to solve. Messy, complex workflows are fertile ground for the deployment of highly capable AI agents that can tackle repetitive, error-prone tasks far more efficiently and with fewer mistakes. The proliferation of SaaS platforms over the last couple of decades has made things even more complex, but many of these platforms will never adapt to the reality of AI. There is massive opportunity in replatforming huge swaths of existing business systems in an AI-first paradigm.

As investors, we have always been excited about great founders solving big problems in a space where they have a “right to an opinion”. In part, this is because developing unique insights into problem-solving for a particular industry remains a very human endeavor. Founders with deep, lived experience in vertical domains are uniquely qualified to reimagine these industries in the context of AI. Finally, one of the primary ways in which AI value will accrue, will be in efficiency and yes, in reduced labor costs. In the near-future world, repetitive manual tasks will increasingly become the domain of AI or AI-powered machines. As scary as this sounds, this will free people to focus on more strategic, creative, and productive tasks and areas that require social interactions, adaptability, and nuance.

Coda

This post is not intended to be a philosophical reflection on AI’s societal impacts, but it’s important to note that we have seen this movie before. In agriculture, horse-powered machines replaced human power in the 1800s, and within several decades, gasoline-powered tractors replaced horses. Steam-engine-powered factories, mines and railroads supercharged the first industrial revolution that employed the tidal wave of humans moving from rural areas to cities. In recent times, automation in factories has displaced some workers, but also created entirely new job categories, increased output, and improved product quality. It is likely the emergence of AI will follow a similar trajectory, and really smart people are getting better at quantifying these impacts— both positive and negative.

Whether you are on Team AI or not, this evolution is in full swing, and the scope of change and economic opportunity may be a once-in-a-lifetime occurrence. Intractable problems that seemed impossible to tackle are now solvable. The capital and time required to build software has never been lower and continues to drop. And even industries and markets deemed too small for robust technology investment will enjoy the benefits of a step-function change in productivity and efficiency.

We live in interesting times, indeed. Buckle up.