Post
Open Source x AI
Much has been said in the past year about the impact of LLMs on software development and developer productivity. But to quickly recap using one of the more well-known tools, GitHub has published several helpful studies on how their AI coding assistant (Copilot) has helped developers:
- 88% reported being more productive, 96% were faster with repetitive tasks, 77% spent less time searching for information.
- 60-75% of users felt less frustrated when coding and 73% stated it helped them stay in a flow state.
- GitHub asked 95 developers to write a web server in JavaScript. 45 used GitHub Copilot and finished in just over 1 hour. 50 did not use Copilot and finished in 2hrs and 40 minutes. And this was before GPT-4 was released.
There are several other studies finding similar productivity boosts. For example, McKinsey found that generative AI helps developers reduce time to complete code documentation by 45-50%, code generation by 35-45%, code refactoring by 20-30%, and more high complexity tasks by 10%. In 2023, Copilot alone suggested and had implemented by developers, over a billion lines of code – “Out there, running inside computers, is code generated by a stochastic parrot.” None of this comes as a surprise at this point. However, numerous other phenomena are converging, which suggests several potential outcomes for the software market. The most prominent of these phenomena include:
- In our year-end writeup we discussed the maturation of SaaS and cloud megatrends. As these technologies become increasingly pervasive, they have set a new standard for software delivery and infrastructure scalability. This maturation has led to a more competitive SaaS landscape, with startups and incumbents competing for market share. As we have all seen, it has never been easier to get started building a software product and never been hard to build a moat.
- Tech companies large and small are more cost conscious than ever. Many have drastically cut burn by reducing headcount and/or reducing software vendors. There were over 260k tech employees laid off in 2023 and as we enter the second half of January, layoffs have continued.
- Cloud costs are top of mind for enterprises across the board, causing some to repatriate their data. While the cloud offers scalability and flexibility, rising costs have led many organizations to reevaluate their cloud strategies, repatriating their data and workloads to on-prem environments or exploring hybrid cloud solutions.
- Generative AI is sparking a new software paradigm, both in how software is built and how it is experienced (we’ve discussed new forms of computer-human interaction previously). These tools harness the power of LLMs to automate various aspects of the software development lifecycle, bringing about significant improvements in efficiency and productivity for developers. And, many are betting on AI-driven applications being capable of understanding individual user preferences, context, and behavior, leading to more intuitive and hyper-personalized experiences.
- The tension between open vs. closed source AI. While OpenAI, Anthropic, Inflection, Cohere, Adept, and others are building closed source models, Meta, Hugging Face, Stability AI and many more are building towards an open-source AI ecosystem. Like any other application or infrastructure, there are both pros and cons for each path, and this conflict is central to the future of AI development and deployment. While OpenAI builds towards what would be a “closed source AGI,” Mark Zuckerberg stated just this week that Meta is all in on developing open-source AGI.
Chamath Palihapitiya recently announced he was starting an incubator, 8090. The idea is to build “an 80% feature complete version at a 90% discount… using AI and offshoring to make this happen.” In the context of the above trends a debate has begun to bubble up around what AI-driven development of OSS could mean for different pockets of vertical SaaS. Namely, if the vertical SaaS market has matured, if differentiation is harder and harder to come by other than price, could open-source products built with AI tooling offer a viable alternative. Our view is that while a new software paradigm is certainly emerging here, there are some things that do not change, and we can look back historically at why open-source has succeeded or failed in beating out closed source competition as an indicator for which pockets of vertical SaaS are most at risk of facing legitimate competition from AI-enabled open-source software. The other challenge of course is that just as open-source projects developers will take advantage of AI tooling, so too will closed-source competitors. Or as Jeff Bezos once remarked:
I very frequently get the question: ‘What’s going to change in the next 10 years?’ And that is a very interesting question; it’s a very common one. I almost never get the question: ‘What’s not going to change in the next 10 years?’ And I submit to you that that second question is actually the more important of the two — because you can build a business strategy around the things that are stable in time.
Open-source code is 80% of software code, more than 90% of developers rely on open-source components, and 96% of applications have at least one open-source component. Linux, Git, Apache Hadoop, Grafana, Confluent, MongoDB, Hashicorp, TensorFlow, and the list goes on and on. It’s hard to overstate how critical this software is for developers and technical users. There are also reasons why historically we haven’t seen open-source application software become as prominent for non-technical business users or consumers in specific industry verticals such as fintech. These products require complex business logic and integrations, demand strict compliance and security standards which closed source software may prioritize and demonstrate more effectively, more quickly respond to, and build new features that align with demands from its users, and come with specialized expertise and support as a default (plus sales efforts, partnerships, etc). Additionally, anyone who has tried open-source business applications will notice that the UI/UX is often lacking (a common complaint). Lastly, something else we’ve written about previously is cyber risks – 82% of open-source software components are “inherently risky” due to vulnerabilities or other issues.
The question is whether a combination of open source with generative AI driven development bridge these gaps. And specifically, whether it will deliver a better product than the closed source competition when they too will be building with the same AI tooling. Or at least, a product good enough at 80% feature complete and at a 90% pricing discount (if possible) such that it takes meaningful share. Unlike vertical SaaS application software that requires industry-specific expertise and compliance standards, horizontal SaaS tools address more general needs (one recent example is Cal.com), which we think makes them a much bigger target for AI-driven open-source development. This is of course all use case specific – there are some verticals where we think unique circumstances lend themselves to open-source products (physical security monitoring software is one such example).
We are very optimistic about how vertical SaaS products will evolve alongside AI-first software development. But even as models become better and better, there are still many hurdles before “AI software development” moves beyond its current role as a productivity booster into end-to-end autonomous software development. The emergence of AI-driven open-source development, as suggested by Chamath incubator raises interesting questions about evolving opportunities in across a crowded SaaS landscape. While a new software paradigm is emerging along with what we expect to be reimagined application software for many industries, historical trends suggest that open-source application software for business users will face challenges in winning significant market share.