Diving deep on AI for software development
Amid that disruption, the job for venture investors is to pick both the sectors and companies with the greatest potential to survive and thrive. For example, we believe AI for software development presents a significant opportunity, potentially enhancing developers' productivity by automating code writing and mundane tasks. This could lead to an increase in both the quantity and quality of software (while also displacing some high-priced software developers). However, with more than 10 startups and a major industry player in the field, identifying the potential winner is essential. This is particularly true given that this space has more than US$10 billion in private market valuations despite no players generating more than US$50 million in run rate revenues (to our knowledge).
Within many categories of AI, from coding to language translation to content creation, investors also consider the possibility that entire sectors could be subsumed by the leading model builders, as those models keep improving and becoming more expert across domains. For example, could OpenAI’s models become so good as they approach (or achieve) Artificial General Intelligence that OpenAI is the best AI model for software development, rather than a vertical application company? We currently think tremendous value will accrue to specialized companies that do jobs exceptionally well, but it’s an important question to which we keep coming back.
Are there still moats in software?
Sticking with AI for software development as an example, in a world where developers are less scarce and great software can be built with little friction, does software become more commoditized and less defensible?
We believe a key takeaway for software defensibility in the AI era is that data advantages and deep customer affinity seem to be compounding. The AI era may push founders to keep their businesses more secretive for longer, while trying to build market leadership and moats through partnerships and data amassed from customer relationships. For instance, a cybersecurity company we’ve partnered with is an interesting example of this, having trained models on customer data since 2018. They’ve built a data moat among over 2,000 customers that could make a new startup struggle to compete, even if they can leverage AI to quickly build exceptional software.
In our view, superseding even data moats, the human aspects of software businesses could become more critical than ever in a world where software can be developed and empowered by AI. The differentiation of your software product may mean less than your go-to-market efficiency and execution. Excellence in sales and marketing has been a critical factor that has made the best startups successful over many decades. As such, our investing frameworks for this new paradigm may not change that significantly. The strongest management teams tend to out-execute (particularly in go-to-market). This can create distance from competitors and help leaders compound at high rates for extended periods of time, particularly when fueled by AI.
Software is a beautiful business model, in our view, because companies can make something once and sell it many times for very little marginal cost (high gross margins). High operational expenses, namely sales and marketing and research and development, have historically proven to be the largest impediments to strong operating margins and profitability for many software companies that failed to reach massive (i.e., US$1B+ revenue) scale. Promisingly, we can now envision a future where software companies can dramatically compress operational expenses by using AI to automate big chunks of sales and marketing and research and development. They would likely still need high-powered account executives and software developers, just far fewer of them that are supercharged by technology. We believe AI may make the sector’s great business model even better.
Bottom line: The outlook for AI startups
The pace of change, model improvement, and overall innovation in AI make the startup winners hard to predict. The same dynamics make us optimistic that as in prior periods of dislocation, huge AI-native companies are being built. There are clear winners in this first wave of growth, but, in our view, it’s just the beginning, and it could take time for applications to take hold. Just look back to VC industry returns for 1999 – 2002, as the foundations of the internet and later mobile waves were being built. Hint: Those vintages barely broke even in aggregate, yet infrastructure investments during this period paved the path for huge returns from applications in the ensuing decades.
Given the creative, human-like nature of AI, far more of the economy is now in technology’s crosshairs, which should bode well for productivity as leaders emerge. Existing software companies that fail to deliver high ROI would have been disrupted regardless, in our view, but AI is a likely accelerant. Moreover, AI may quickly redefine what we mean by high ROI, increasingly adding software use cases to the menu.