Archive no. MY-0007
Kite
The best-known AI code-completion startup of the 2010s, which shut down in 2022 with a founder's post-mortem arguing it was ten years too early and that developers would not pay for speed.
Original intent · 初衷
Kite set out to deliver AI-assisted programming: the founder describes the goal as a 10× improvement in how developers write code, built on models trained specifically for code rather than general text. The company's own framing of what went wrong — that the technology was not ready and that the product did not monetize — is part of the record it left behind.
Timeline · 时间线
-
2014 · year-level
Kite starts as a startup using AI to help developers write code.
-
2019 · year-level
The company reaches product-market fit, five years after starting.
-
2022-11-16 · date-level
Kite publishes its farewell post: the team has stopped working on Kite and no longer supports the software.
Dates are shown at the precision the sources provide; nothing is rounded up to a full date.
What happened · 后来发生了什么
Kite built its engineering team first, then the product, and did not reach product-market fit until 2019, five years in. It grew to 500,000 monthly-active developers with almost no marketing spend, and then failed to generate revenue: those developers would not pay. The founder's diagnosis is that individual developers do not pay for tools, and that engineering managers only buy discrete new capabilities, so making developers 18% faster did not resonate strongly enough. The team explored a pivot to code search before concluding it was too tired to pursue it, and shut the product down instead.
Why it stopped · 停止原因
The founder names two failures rather than one. First, Kite was too early: the state of the art for machine learning on code was not good enough for the 10× improvement that would have broken through, and building production-quality code synthesis may cost over $100 million. Second, and independent of the first, the business did not monetize — 500,000 monthly-active developers did not pay, and managers would not buy a general speed-up. Neither reason is an outside speculation; both are the company's own retrospective judgement.
Curator's reading · 策展人分析
Facts cite a source. Author statements are the makers' own words. Curator analysis and hypotheses are marked as interpretation.
Editor's reading: the founder's explanation is a product and business analysis, not a market verdict on AI coding itself. He argues the direction was right and early, and that the company could still have built a business without a 10× gain if it had found something developers or their managers would pay for. Kite's archive value is that it is the category's origin story told by the people who ran it.
What remains · 留下了什么
No leftover artifact is recorded for this entry: no public repository, demo or write-up we could verify.
What the makers learned · 制作者总结
The founder states that Kite stopped working on the product and is no longer supporting the software.
The founder's own assessment is that the company was ten or more years too early, that machine learning on code was not good enough for the improvement required, and that it separately failed to monetize.
Kite reports growing to 500,000 monthly-active developers with almost no marketing spend, and states that those developers would not pay to use the product.
After exploring a pivot to code search, the team decided it was too tired to pursue it and chose a soft landing instead.
The farewell post's visible byline prints November 16, 2021, while the page's own text refers to late 2022 and its structured metadata dates the post 2022-11-16. This entry records the 2022 date and flags the discrepancy rather than picking silently.
Two product judgements come out of this case. The first is about willingness to pay in developer tools: a large free audience is not evidence of a business, and a general productivity gain of roughly 18% was not a purchase trigger even for the people who used the product daily. The second is about timing and platform gravity: Kite's own post-mortem points at GitHub Copilot, built by the platform owner with a model partner, as the effort that picked up the direction — the startup that defined the category watched a distribution owner take it. Kite is also a rare honest artefact: most failed AI products leave no explanation of why, and this one documents the sequencing decision — team, then product, then distribution, then monetization — that made the failure legible in hindsight.
If someone tried again · 重做问题
Kite monetized last, after team, product and distribution. If it had tested willingness to pay earlier — selling a narrow, discrete capability to engineering managers rather than a general speed-up to individual developers — would the company have discovered the business was unviable before spending seven years on the technology?
Sources and verification · 来源与核验
- Status evidence
- [kite-farewell-post]
- Added
- 2026-09-13
- Entry updated
- 2026-09-13
- Last verified
- 2026-09-13
- Project period
- 2014 → 2022-11-16
Supports: Kite's founder states the company stopped working on Kite and no longer supports the software, and explains why the business failed.
Supports: Second-party coverage corroborating the shutdown; it carries no independent state claim.