The Value of AI Programming Assistants, and Where Cursor May Go Next

Jiang Wu · 2024-09-08

*Originally published in Chinese on September 8, 2024.*

Thinking through programming assistants

Cursor, an AI code editor, has been popular on Twitter for a long time, and I have followed it for a while. At first only programmers I follow were using it, and now it seems to be getting more and more popular. I am not a programmer, but I know the group fairly well and gave the tool a quick try. I start from the tool itself and look at three things: its core value, its use cases, and its barriers to value.

Breaking down the value of programming assistants

The questions are what problem the tool solves, what its use cases are, and what its core barrier to value is. Measured against those, a programming assistant can be broken down step by step.

Do programmers really need it?

Is an automatic programming tool a real need for programmers, something that solves a problem for them?

Of course it is. Most of the process of programming is repetitive work. The core value lies in the solution and the act of creating, and getting code written and running is mostly routine, tedious, repetitive work. From that angle, a programming assistant frees up programmers' time and improves their efficiency.

Where programming tools are used

For freelancers and programmers who are not strictly constrained, the working environment is fairly free. They use the code tools they are used to, on the platforms and devices they are used to, and there is almost no problem. For a fully independent individual, a programming tool can fully release value and raise productivity. Some settings have requirements, though, such as large companies, particular industries, and companies with high confidentiality demands. Their requirements and review of the working environment are extremely strict. They restrict the office environment, the programming setting, and the tools, and these form fairly strict barriers to use. In the current real environment, many large companies still have strict requirements on this. Some have rules on whether and how such tools can be used and some are vague. But if the company has not authorized a product like this on the machines its people work on in bulk, an individual may well still have a problem.

The barrier to value

What would the barrier to value be for a programming tool?

Code that completes itself, that understands your intent, your path to implementing the product, even your plan and requirements for the whole product, are all barriers to value. Through understanding the product and how it will be built, the tool delivers value in small differences, and often does better than a person would. That is a very strong barrier to value.

The biggest problem facing this barrier is that, in the product development system of the future, the core capability may not be a programming assistant. It leans toward lighter one-stop services that wrap all the code and functions behind an interface for users to call, and users may not need complex programming at all. So the choice of where the value sits may need some time to observe. This is not a problem with the value of programming assistants. It is that the product or the industry may end up treating basic programming ability as a light need, and that future can be seen coming soon.

The market today and what could happen

With the product and value questions covered, here is the current market in China.

Model companies are all entering

Nearly every model company has launched a similar product, without exception. Completing software code is relatively easier than completing human language, because code follows set rules and systems and the model only has to understand the rules and continue the text. The barrier to developing the product is therefore not that high. A standalone tool product may face a larger challenge, since it is one link that many model products already cover.

Limits set by large companies

Second, large Chinese companies of any scale, and companies with high information security requirements, generally have strict rules for the programming environment. On this point programmers do not necessarily get to choose freely, so who finally pays for the product depends on the environment of each system. If the company ends up paying, it may choose local deployment.

Opportunities and challenges for cloud providers

Third, much code deployment and many services need not be handled locally, and the tool environment is still dominated by some common tools (including the big companies and the editors). Cloud providers may well be an important distribution channel for products on this dimension, and I would not rule out that they launch similar products themselves. The major Chinese cloud providers are all backed by companies with both model capability and tool capability. Abroad, cloud providers already have them, and both Azure and AWS officially offer similar products. So a standalone product that wants to become a system needs more than product capability. It needs something with more of a barrier.

Where Cursor goes next

Now I can guess at Cursor's next step.

Home · All articles · Topics · 简体中文