Product reviews, deals and the latest tech news

Can generative AI for code succeed after Kite’s demise?

Last month, Kite, a company that was working on an artificial intelligence-powered coding helper, unexpectedly ceased operations. Kite’s founder Adam Smith wrote in a postmortem blog post that the company had trouble making ends meet despite receiving tens of millions of dollars in venture capital funding. The company had trouble establishing a product-market fit due to engineering challenges.

Unfortunately, as Smith put it, “the technology is not there yet,” thus they were “10+ years too early to market” with their vision of AI-assisted programming. It took too long to realise that our product was not profitable.

Kite’s demise is discouraging news for the many other businesses looking to market generative AI for programming. Perhaps the most well-known instance is Copilot, a $10/month code-generation service created by GitHub and OpenAI. However, Smith emphasises that despite Copilot’s promising early results, the project still has “a long way to go” and might end up costing more than $100 million to develop a “production-quality” tool that can consistently synthesise code.

TechCrunch spoke to businesses building AI systems for coding, such as Tabnine and DeepCode (bought by Snyk in2020), to get a feel of the difficulties that lie ahead for companies in the generative code market. Similar to Copilot, Tabnine’s service may anticipate and recommend the next lines of code to be written depending on the current context and syntax. In contrast, DeepCode use AI to flag coding errors in real time.

Tabnine’s chief executive officer, Dror Weiss, was candid about the obstacles he sees to widespread usage of code-synthesizing systems: the AI, the user experience, and monetization.