Granola is the AI notepad for people in back-to-back meetings. It quietly transcribes your calls, blends the transcript with your own notes, and turns everything into clean, searchable summaries.

Granola is trusted by Vercel, Ramp and Brex as their AI meetings notepad of choice.

And they ship fast. Most of their code is now written with the help of AI. By going all-in on these tools, the team stays immersed in the medium, constantly navigating the same AI strengths and hallucinations their users face.

The question became how to keep standards high when AI turns every engineer into the manager of a small “org of agents,” instead of a traditional individual contributor. That is where cubic comes in.

The bottleneck: code review, not code writing

Mathew has spent over a decade inside large-scale systems, including 11 years at Google and helping start the Uber Eats engineering office in New York. Now at Granola, his job is to design systems the team can “build once and not have to keep fiddling with.”

To do that, the team is all-in on AI. They lean heavily on tools like Cursor and run multiple worktrees with AI agents in parallel. They even have a Slack channel to swap prompts and patterns.

But as AI output ramped up, code review started to drag:

• Senior engineers were pulled into more and more reviews.

• PR queues grew, and context switching got painful.

• Human review time was still going to missing tests, edge cases, and unsafe patterns.

At the same time, the role of engineers was shifting.

"People’s roles are moving away from classic IC work to managing fleets of agents. You have to think about how to split up the work and how to make sure those agents are doing the right thing."

— Mathew Cowan, Chief Architect at Granola

So Granola went looking for AI code review. They needed something that would sit next to Cursor and act like a tireless senior engineer, not a noisy bot.

Why Granola chose cubic

Granola tried other AI review tools, including CodeRabbit, but hit the usual roadblocks: hallucinated issues, "vibes-based" comments, and painful configuration. “Tools that create noise get muted,” Mathew notes. “Once that happens, you’ve lost the safety net.”

They switched to cubic because it focused on signal.

"cubic feels like a senior engineer who only speaks up when it matters.”

— Mathew Cowan, Chief Architect at Granola

The team interacts with cubic directly in GitHub. They rely on the auto-updating PR summaries and the consolidated list of issues. It removes the friction between PR opened and PR merged.

Catching the bugs humans miss

The clearest ROI wasn't speed - it was security.

With AI generating more surface area, the risk of subtle vulnerabilities creeps up. Cubic became the safety net that allowed them to move fast without breaking things.

"cubic has already paid for itself just in the bugs it caught before they ever reached production. It has caught real issues that would have turned into painful post-mortems.”

— Mathew Cowan, Chief Architect at Granola

For a Chief Architect responsible for risk at scale, that reliability is the only metric that counts.

“Knowing cubic is looking at every PR reduces our risk. It genuinely helps me sleep better at night.”

Turning incidents into guardrails

Granola leans heavily on custom rules to enforce lessons learned.

Instead of relying on tribal knowledge or a Slack thread from six months ago, they encode the fix immediately.

"We'll often write a rule the moment we fix something painful, so next time cubic can catch it automatically."

— Mathew Cowan, Chief Architect at Granola

Whether it is deprecating old APIs, adding scrutiny to Terraform changes, or enforcing onboarding standards, they write the rule once and let cubic enforce it forever.

The new workflow

Granola has adapted their entire process for an AI-heavy world.

1. Plan: Humans write RFCs and review architecture.

2. Build: Agents and engineers generate the implementation.

3. Review: cubic handles the low-level review, security checks, and convention enforcement.

"I want developers reviewing the agent's plan, not every single line it generates"

— Mathew Cowan, Chief Architect at Granola

By offloading the review grunt work to cubic, Granola keeps their velocity high without sacrificing the quality of their codebase.

Ready to automate your code review?

Start your free trial or book a time with us to get set up.