PlanetScale 为其客户最敏感的数据管理云数据库工作负载。可靠性就是产品本身，而每一次推送到生产环境的代码变更都必须完美无瑕。随着智能体让代码生成变得廉价且快速，代码审查成为了软件开发生命周期中新的瓶颈。
为了确保正确性并充满信心地将代码交付到生产环境，PlanetScale 采用了 Bugbot 作为专门的智能体审查层。如今，大约 80% 的 Bugbot 评论在合并前得到处理，从而防止问题进入生产环境，并为 PlanetScale 节省了相当于两名全职工程师的审查工作量。
智能体将 SDLC 中的瓶颈向下游转移
为了维护产品可靠性，PlanetScale 工程团队为代码审查设定了不容妥协的标准。“可靠性是我们产品的核心。每一次推送到生产环境的变更都必须完美无瑕，”PlanetScale 软件工程师 Fatih Arslan 说道。
随着编码智能体成为开发工作流程的核心，瓶颈向下游转移，从代码生成转移到了代码审查。“代码变得廉价。现在的瓶颈在于你的代码是否正确，以及你是否理解它的作用，”Arslan 解释道。
代码产出迅速增加，而人工审查能力保持不变。这种不平衡给产品质量带来了压力。为了跟上节奏，PlanetScale 估计需要专门安排两名工程师全职进行代码审查。这种权衡会减少用于产品开发的工程资源，并且随着智能体采用规模的持续扩大，也无法解决长期可靠性挑战。
我们意识到需要 Bugbot 的智能体审查来补充我们现有的流程。否则，很难在质量和正确性方面充满信心地将代码推送到生产环境。
Fatih Arslan
PlanetScale 软件工程师
利用 Bugbot 消除生产环境停机
在检测人工审查员因 PlanetScale 代码库的复杂性和智能体生成代码的数量而遗漏的问题方面，Bugbot 在其他智能体审查工具中脱颖而出。
借助 Bugbot，工程师们开始在开发早期发现并解决可能导致生产环境停机的错误。
Bugbot 与其他工具不同。它能检测出人类审查者根本想不到要去查看的问题。这让我大为震撼。
Fatih Arslan
PlanetScale 软件工程师
与专注于机械正确性的静态分析器和代码检查工具不同，Bugbot 能揭示更深层的语义和逻辑问题，例如：
- 状态同步缺口，导致系统被过早标记为完成
- 逻辑流程变更，阻止了关键代码路径的执行
- 异步控制器交互，未能正确收敛
- 可能触发生产数据库重启的边界情况
Bugbot 总能发现那些可能导致生产环境严重停机的错误，而这些错误人类很难捕捉到。
Fatih Arslan
PlanetScale 软件工程师
PlanetScale 还发现，仅仅提示一个前沿模型来审查代码，并不能始终如一地发现 Bugbot 所识别出的最关键问题。Arslan 表示：“当我使用推理模型并要求它审查分支时，它找不到这些问题。正是专门的测试框架和 Bugbot 的构建方式，才带来了天壤之别。”
衡量 Bugbot 审查质量
PlanetScale 使用一个简单的指标来评估 Bugbot：解决率，即衡量在合并时被处理的 Bugbot 识别问题的比例。
目前，在每月审查的超过 2000 个 PR 中，大约 80% 的 Bugbot 评论已被工程师处理。Arslan 表示：“Bugbot 的评论质量一流，并且随着 Bugbot 获得更多上下文，其质量还在不断提升。”
Bugbot 的信噪比非常高。当 Bugbot 对某个 PR 发表评论时，我们知道它正在指出一个我们必须修复的问题。
Fatih Arslan
PlanetScale 软件工程师
Bugbot 现已深度融入 PlanetScale 的工作流程，让工程师们有信心将人工编写和智能体生成的代码安全地发布到生产环境。Arslan 说：“我爱 Bugbot。这就是我的座右铭。”
PlanetScale 现在可以在不牺牲质量的前提下更快地发布软件，同时工程师可以专注于解决复杂的基础设施问题，而不是手动审查智能体生成的每一行代码。
如果我把 Bugbot 从我们的工程团队中拿走，那肯定会引起哗变。
Sam Lambert
PlanetScale 首席执行官
如果你对利用智能体简化代码审查、提升产品可靠性感兴趣，欢迎联系我们的团队，开始试用 Cursor。
更多客户案例
Coinbase 借助 Cursor 将创意到上线的时间缩短了 90%
Wayfair 如何通过 Cursor 将机器学习模型成本降低 90%（两次！）
Faire 借助 Cursor Cloud Agents 将 PR 吞吐量翻倍
查看所有案例 →
PlanetScale manages cloud database workloads for its customers' most sensitive data. Reliability is the product and every code change pushed to production must be flawless. As agents made code generation cheap and fast, code review became the new bottleneck in the software development lifecycle.
To ensure correctness and ship code to production with confidence, PlanetScale adopted Bugbot as a dedicated agentic review layer. Today, roughly 80% of Bugbot comments are addressed before merge time, preventing issues from reaching production and saving PlanetScale the equivalent of two full-time engineers worth of review effort.
Agents shift bottlenecks in the SDLC downstream
To maintain product reliability, the PlanetScale engineering team sets an uncompromising bar for code review. "Reliability is at the core of our product. Every change pushed to production must be flawless," said Fatih Arslan, a software engineer at PlanetScale.
As coding agents became central to development workflows, bottlenecks shifted downstream, from code generation to code review. "Code has become cheap. The bottleneck is now whether your code is correct and whether you understand what it does," explained Arslan.
Code output increased rapidly while human review capacity remained fixed. This imbalance created pressure on product quality. To keep pace, PlanetScale estimated it would need to dedicate two engineers exclusively to code review. That tradeoff would reduce engineering bandwidth for product development without solving the long-term reliability challenge as agent adoption continued to scale.
We realized we need Bugbot's agentic review to complement our existing process. It would otherwise be very difficult to push code to production with confidence in quality and correctness.
Fatih Arslan
Software engineer, PlanetScale
Eliminating production downtime with Bugbot
Bugbot stood out from other agent review tools in detecting issues that human reviewers were missing because of the complexity of PlanetScale's codebase and volume of agent-generated code.
With Bugbot, engineers began catching and resolving bugs that could cause production downtime earlier in development.
Bugbot is different from other tools. It detects issues that as a human reviewer I would never imagine to look at. I was blown away.
Fatih Arslan
Software engineer, PlanetScale
Unlike static analyzers and linters that focus on mechanical correctness, Bugbot surfaces deeper semantic and logical issues like:
- State synchronization gaps where systems are marked complete prematurely
- Logical flow changes that prevent critical code paths from executing
- Asynchronous controller interactions that fail to converge properly
- Edge cases that could trigger restarts across production databases
Bugbot consistently finds bugs that can cause meaningful downtime in production, but are very hard for humans to catch.
Fatih Arslan
Software engineer, PlanetScale
PlanetScale also found that simply prompting a frontier model to review code would not consistently surface the most critical issues Bugbot identifies. "When I use a reasoning model and ask it to review the branch, it doesn't find these issues. It's the specialized harness and the way Bugbot is built that makes all the difference," said Arslan.
Measuring the quality of Bugbot reviews
PlanetScale evaluates Bugbot using a simple metric: resolution rate, which measures the proportion of Bugbot-identified issues that are addressed at merge time.
Roughly 80% of Bugbot comments are now addressed by engineers across more than 2,000 PRs reviewed each month. "Bugbot's comments are top notch and keep improving over time as Bugbot gets more context," said Arslan.
The signal-to-noise ratio for Bugbot is very high. When Bugbot comments on a PR, we know it is highlighting an issue we have to fix.
Fatih Arslan
Software engineer, PlanetScale
Bugbot is now deeply embedded in PlanetScale's workflow and gives engineers the confidence that both human-written and agent-generated code can be shipped safely to production. "I love Bugbot. That's my motto," says Arslan.
PlanetScale can now ship software more quickly without sacrificing quality, while engineers can focus on solving complex infrastructure problems rather than manually reviewing every line produced by agents.
If I took Bugbot away from our engineering team, there would be a mutiny.
Sam Lambert
CEO, PlanetScale
If you're excited about streamlining code review and enhancing product reliability with agents, please reach out to our team to get started with a Cursor trial.
More customer stories
Coinbase reduces time from idea to production by 90% with Cursor
How Wayfair cut ML model costs by 90% (twice!) with Cursor
Faire doubles PR throughput with Cursor Cloud Agents
View all stories →