档案库 · 开发与企业工具 · 产品决策 · 2022
Litebulb(YC W22)押注真代码库编程面试会击败LeetCode
YC支持的Litebulb将编程面试自动化,让候选人在真实代码库上做真实工作,而非LeetCode题目,并于2022年3月推出,已有公司付费。
Litebulb
做的是什么生意
SaaS platform for employers: candidates get a Dockerized real codebase, build a feature per specs, submit a pull request, and Litebulb runs tests, load and static analysis, then returns a hiring report.
起因
Gary Lin, who had worked at 11 companies and been through well over a hundred interview funnels, founded Litebulb after concluding that puzzle screening tested the wrong skill while interviews drained engineer time. The company went through the YC W22 batch, and by early 2022 it was hiring its own engineers through the product and had paying customers before the public launch.
经过
The Launch HN went up on 2022-03-07 and drew 139 points and 190 comments. The pitch was concrete: interview environments are Dockerized, setup is boiled down to a single make command, candidates code in Codespaces and submit a pull request, and Litebulb runs linter, integration and visual-regression tests, load tests, and complexity checks before compiling a report. Gary Lin positioned it as 'the depth of a service like Karat at the scale and price point of a tool like HackerRank', with a long-term ambition to become 'Webflow for interviews'. His replies named Gumroad, On Deck, Mashgin, Dover, SnapEDA, Evidence.dev, getatlas.io and Okteto as companies running Litebulb interviews. Much of the thread became a debate about whether automated assessment disrespects candidates, whether code metrics can judge people, and whether cheating is preventable; commenters also quoted pricing from $600/month up to enterprise tiers. Gary answered that Litebulb belongs mid-funnel with human screens before and after, and that a V2 scorecard would expose raw data instead of labels like 'strong junior'.
还没有结局,它还在跑。
背景
Litebulb是一家YC W22初创公司,为远程团队自动化编程面试。创始人Gary Lin是一名软件工程师,曾在11家公司工作,经历过上百个面试流程,他认为LeetCode式谜题测试了错误的技能,而构建和进行技术面试让工程师无法真正工作。他的赌注是,获胜的工具将自动化漏斗后期阶段,Karat用人工服务,而HackerRank则完全跳过。
这个产品让候选人做真实工作:访问包含数据库、服务器和客户端的现有代码库,用一条make命令启动Docker环境,在基于浏览器的Codespaces中编码,以GitHub pull request形式提交。Litebulb运行测试、linter、视觉回归、负载和复杂度分析,并生成雇主报告,但在测试版期间仍有人工验证结果。2022-03-07在Hacker News上发布,帖子获得139分和190条评论,创始人在回复中点名Gumroad、On Deck、Mashgin、Dover、SnapEDA、Evidence.dev、getatlas.io和Okteto为使用者。
反应两极分化。讨厌LeetCode刷题的招聘经理和候选人称赞真实代码库的方法;其他人则批评自动化评估缺乏人性化,说数字指标制造了虚假的客观性。Gary回应称Litebulb不是预筛选工具,建议前后安排人工电话,并且正将记分卡从资历标签转向原始数据。发布时仍处于测试阶段,用24小时周转人工验证结果,并计划推出数据科学面试、候选人端平台和ATS集成。
这件事要成立,得有什么
- 工程师时间不仅是技术招聘的最大成本:在职工程师需参与构建、运行和评分面试,Gary提到团队每周花费两小时或多次现场面试占比不小。
- 市场存在空白:漏斗早期筛选用谜题,Karat提供人工高成本,因此自动化后期漏斗产品以工具价位可差异化。
- 真实代码库任务规避了谜题的失败模式:没有可复制的标准答案,git历史diff标记复制解决方案,准备意味着成为更好的开发者,而非LeetCode解题者。
- 发布争论暴露了核心风险:候选人和招聘经理不信任自动化判断,而像“高级初级”的标签损害客观性,推动Litebulab转向原始指标。
可借鉴之处
针对人们讨厌的现状进行定位能吸引眼球,但产品仍需证明其信号对双方都有意义;Litebulb的发布战是关于信任自动化,而非功能。
后续进展
截至2022-03-07,Litebulb仍处于早期状态:测试版至少再持续三个月,24小时人工验证每次面试结果,但已有付费客户,包括Gumroad、On Deck、Mashgin、Dover、SnapEDA、Evidence.dev、getatlas.io和Okteto。路线图包括V2记分卡(暴露原始数据而非资历标签)、雇主自助分析、2022年晚些时候推出候选人练习平台、数据科学面试,以及ATS集成(如Greenhouse和Lever)。Gary Lin在HN上表示,最大的风险是执行,而非市场规模。
资料来源
发现哪里写错了?告诉我们。
轮到你了
你刚读完一家。说说你在做什么,看看谁在赌同一件事。
免费账号 · 3 次免费提问 · 不用绑卡