/WRITING/no-moats
There are no moats in AI software developer tools
There are no moats in AI software developer tools.
Traditionally, software development has been a land of moats. Existing tools, like programming languages, cloud service providers, and even editors, had strong, defendable positions in the market. At the individual level, changing to, say, a different programming language has been easy - all you need to do is install an additional CLI tool and editor extension. However, at the corporate level, things have never been that straightforward. Tool selections made by the company affected internal procedures, hiring, company structure, and profitability. Companies occasionally tried to make big changes to their tooling, like moving their entire infrastructure to a different cloud provider or adding a whole new technology stack to their huge web apps (just random examples ;-) ), but those projects always took a long time, cost a lot of money, and often failed. Furthermore, software developers used to specialize in specific technologies and were highly opinionated about their tooling choices on an individual level. I’ve seen this firsthand as someone who used to promote an “underdog” programming language1 on .NET and make tools2 for an IDE that wasn’t typical in the .NET world.
That being said, it looks like this resistance to change is no longer there. The situation has entirely changed in the last few years, when AI tools have taken center stage. In just 5 years since introducing the original GitHub Copilot, we went from Copilot in-editor tooling to various VSCode forks to various AI agent harnesses. And even in this final category, Claude Code was the dominant force until about eight months ago. Today, people are moving to Codex or various other OSS alternative harnesses. This is a thrilling and scary time for toolmakers. On the one hand, we have a massive opportunity to reach new users that we have never had before. However, it also means that people will gladly switch to something else if they think it is better. Now, let’s talk about why this happens and why people are so willing to give up the tool they’re using and start using something else.
First and foremost, AI tools are made to simply adjust to your context. Whatever tool you choose, the underlying mechanism is broadly the same; ultimately, it’s just a series of calls to LLM. And the most magical thing about LLMs is that they can change into anything you give them. Do you want to use them to write a fantasy novel? Sure, no problem. Do you want to code in some esoteric programming language? Sounds great. Do you want to optimize some benchmark results? Perfe… actually, maybe let’s not do this3. Since the new tool will simply adjust to your current workflows with little modification, switching is incredibly inexpensive. This is particularly true today, as there aren’t many differences between the most advanced models offered by various providers.
The next important aspect is that this is a very new field. We started only 5 years ago. And the past few years have been a period of significant innovation and investment in both the AI models and the tooling, resulting in more frequent and significant changes than anticipated. In a mere five years, we have transitioned from a very limited in-editor experience to agent harnesses that operate continuously for several days without any user input, enabling the creation of extremely complex projects from scratch. In contrast, traditional tools have been well established and were only incremental changes. Similarly, there haven’t been considerable differences between existing tools. For example, we could argue about which cloud provider is better or worse, but all the major ones have about the same level of functionality.
This is somewhat related to the fact that we have disrupted the conventional adoption model with AI tools. Typically, there is an adaptation curve for any business product, meaning that at first it is only utilized by enthusiasts and early adopters before becoming more widely used. It basically means that any product that manages to cross the chasm4 and reach mainstream has already been fairly established and iterated on. The situation has changed with AI tools, however. Even the first AI developer tool, GitHub Copilot, was widely adopted very fast. Additionally, all the well-established tooling vendors, including Microsoft, Google, AWS, JetBrains, and others, were actively promoting their AI products. An environment entirely different from the traditional tooling ecosystem was created as a result of persuading users who were historically more conservative (in terms of adopting new technologies) to adopt relatively unproven technologies.
Creating this environment has an indirect impact on existing sales networks and partnerships, which is one of the most important traditional moats. In the old days, if you were a Microsoft customer with thousands of Office suite seats, you would probably pick Azure as your cloud provider. Your Microsoft sales representative would likely be able to negotiate better support access and some sort of discount for you. Vertical integrations had a lot of value. For example, if you use Azure, you should use .NET because it will have the best support, and Visual Studio because it will have the best support for C#, and so on. All major corporations were accustomed to the second mover advantage; they were not required to be the most innovative; they could observe the market, identify successful products, replicate them, and sell them to their existing clientele through extensive sales networks. Vertical integrations and old-fashioned ways of selling don’t work in the new world. And because of the conditions we’ve already talked about, the use of AI tools is largely determined by the opinions of developers rather than by business decisions made by companies. Although they were already important, social media and community enthusiasm are now the main ways to shape developers’ thoughts and interests. They are also a major factor in the adoption of AI tools because people are eager to try any new tool that creates excitement on social media.
So, now after we’ve discussed in far too many paragraphs why traditional moats don’t exist in this new era of software developer tools, let’s briefly discuss what we can do to “win” in this new environment. How can we promote our tools? How to ensure they get users and keep them around?
Furthermore, the response is fairly straightforward and comprises two primary ideas. Firstly, you need to remember this is still a young space, where innovation matters. People are willing to quickly adopt new tools because of the environment we’ve created, but you have to be on the cutting edge. Particularly for a well-known business like Microsoft, you can’t rely on your standard operating procedure. If you attempt a typical second mover strategy, you will soon discover that by the time you have created whatever you believe is the best thing to imitate, people have already switched to something else. And the product is already out of date by the time your version comes out. Secondly, it may sound simple, but all you need to do is give people fun and enjoyable experiences. In the past, people were putting up with certain issues or restrictions because switching was so expensive. They may have been acceptable if your product received small, incremental updates every quarter. This has changed, though. The reputation of a tool vendor is influenced by every small detail. And the hardest currency in this world is users trust and reputation. If you are perceived as a slow-moving elephant with a dull product, it will be irrelevant how many corporate blog posts you post claiming that your product looks good in benchmarks. You’ve already lost.
In this post, I have discussed the unique adaptation patterns and massive innovation wave that have resulted in an ecosystem that lacks traditional moats. The market for AI developer tools is always up for disruption, which poses serious problems for well-established vendors who are accustomed to using their second mover advantage to target conservatives and pragmatists4. Rather, the field is dominated by leaner startups that are capable of innovating, generating excitement on social media, and providing exceptional user experiences.