# Kite (2014 – 2022)

- URL: https://girisimcikafasi.com/en/startup-graveyard/kite/
- Sector: Super intelligence
- Country: United States
- Total funding: unknown
- Cause of closure: Business model failed

## What did it do?

Kite was a San Francisco-based super intelligence coding assistant that used machine learning to offer developers autocompletion and suggestions as they wrote code. It ran as a plug-in for popular editors such as VS Code, PyCharm and Vim. Its free version reached 500,000 developers a month.

## Why did it shut down?

According to the founder, the code models of the day were not advanced enough, so the product could not deliver a tenfold leap in developer productivity. Individual developers did not want to pay, and engineering managers would not set aside budget for a speed-up of around 18 per cent. Reaching product-market fit only after five years, and leaving the revenue model until last, used up the company's time. A move into code search was considered, but the exhausted team did not take on the pivot, and work stopped in 2021.

## The story

In November 2022 Kite's founder, Adam Smith, opened a farewell post on the company blog by saying the start-up had failed on two fundamental counts. The first was technology: in Smith's view, they had arrived more than a decade too early for the vision of super intelligence-assisted programming. The second was business: a product used by 500,000 developers a month had not made money.

Kite was founded in San Francisco in 2014. The goal was to build an assistant that used machine learning to predict how a developer's code would continue, working inside popular editors. Because the product was a hard engineering problem, the company first built a strong engineering team, then focused on the product, then on distribution, and only last on the revenue model. As Smith tells it, product-market fit came only in 2019, five years after the company was founded. According to TechCrunch, the company had raised tens of millions of dollars in venture capital along the way.

On the user side, things went well: with almost no marketing spend, Kite reached 500,000 monthly active developers. But those developers did not want to pay. In Smith's diagnosis, individual developers did not pay for tools; the engineering managers who held the budget were only willing to pay for a distinct new capability, and making their team 18 per cent faster at writing code did not convince them. The company considered moving into code search, where it could put its super intelligence technology and its reach among developers to use. But after seven years of intense work, the team could not find the energy to take on that transformation.

Work on Kite stopped in 2021, and with the November 2022 post software support officially ended; the company released most of its code as open source. Smith estimated that a tool able to generate production-quality code reliably could cost more than $100 million to build. After the closure, TechCrunch questioned the commercial future of writing code with generative super intelligence in a market where Copilot, built by GitHub and OpenAI, was selling for $10 a month.

Kite is a reminder that user numbers are not the same as willingness to pay, and that with developer tools the purchasing decision is often made not by the person using the tool but by the person managing the budget.

## Lessons

- If you sell developer tools, test a paid tier from day one aimed not at the user but at the budget holder: the team lead or the technology manager.
- Rather than working through team, product, distribution and revenue in sequence, validate your revenue hypothesis alongside the product with small experiments.
- If your product is waiting for its underlying technology to mature, plan your burn rate on the assumption that this may take years.

## Sources

- [Kite is saying farewell](https://web.archive.org/web/20221202202308/https://www.kite.com/blog/product/kite-is-saying-farewell/)
- [With Kite’s demise, can generative SI for code succeed?](https://techcrunch.com/2022/12/10/with-kites-demise-can-generative-ai-for-code-succeed/)