"I don't know what sort of magic you're doing to understand the ramifications of things that affect other files, but it's super impressive."
- Gergő Móricz, Founding Engineer, Firecrawl

Firecrawl provides web data infrastructure for thousands of AI applications, including Shopify's AI features and Zapier's automated workflows.

With over 51,000 GitHub stars and a complex codebase across TypeScript, Rust, and Go, their engineering team was spending too much time on pull requests while critical bugs slipped through.

After implementing cubic's AI code review, they cut manual review time by 70% and started catching serious bugs before they reached production.

The challenge: Too much code, not enough reviewers

As Firecrawl grew rapidly, code reviews became a major bottleneck:

"I'm doing 10 to 15 PRs a day. It's an incredible amount of review load to put on someone."
- Gergő Móricz, Founding Engineer, Firecrawl

The real problems:

• Production was breaking from unreviewed PRs


• While 90% of PRs were fine, the critical 10% with real issues weren't getting caught

• External contributions needed triple-checking: "I look twice, three times at everything to make sure nothing breaks"


• One engineer reviewing 10-15 PRs daily while still trying to ship their own code

The solution: cubic's AI code review

"The difference in the types of feedback that came back from cubic was immediately shocking—in the best way possible."
- Gergő Móricz, Founding Engineer, Firecrawl

How cubic helped:

• Understands what developers are trying to do and works with their intent

• Catches architectural issues across multiple files that human reviewers might miss

• Handles their complex tech stack seamlessly (TypeScript, Rust, and C), even catching tricky bugs at language boundaries

Why cubic over Copilot

For Firecrawl, the difference was clear:

"Copilot will spam me with 'Hey, this is bad. Hey, this is bad.' It assumes I don't know what I'm doing. But cubic understands that I probably intended to do this, and just makes sure I know the ramifications."
- Gergő Móricz, Founding Engineer, Firecrawl

What this means in practice:

• Respects your expertise and understands trade-offs

• Focuses on real production risks, not nitpicks

• Catches cross-file issues that even senior engineers miss

Real example: The Redis bug that almost shipped

While building a batch billing queue in Redis, cubic caught a subtle but critical bug:

"We were implementing a batch billing queue in Redis. Cubic warned: 'If this is undefined, the downstream consumer will throw an error and this batch will be ignored.' I thought, 'No, that's not going to be undefined.' Then the bug happened. I looked back at the PR and was like, 'God damn it. Cubic was right.'"

The results

Before cubic:

• 10-15 PRs needed CTO review daily

After cubic:

• PR reviews reduced: From 10 to 3 requiring CTO review

• Trusted PRs: 7 out of 10 PRs completely trusted to cubic

Why it works

"Cubic is very good at pointing out when my assumptions are wrong. That's what we were using human reviews for—somebody else on the team might know something you don't. Cubic understands the entire architecture so well that it highlights things other engineers would catch."
- Gergő Móricz, Founding Engineer, Firecrawl

For engineering teams like Firecrawl:

• 70% less time on manual reviews


• External contributors fix their own issues before review

• Works directly in GitHub (no new tools to learn)

• Confidence to accept more open-source contributions

Firecrawl (51,000+ GitHub stars) provides the web data API powering AI applications at Shopify, Zapier, and Replit. Learn more at firecrawl.dev