Quality assurance · Production

cubic reduces false positives by 51% by orchestrating their multi-agent system with Inngest

The problem

cubic's initial single-agent AI code review system produced an influx of low-value comments and false positives, while the multi-step agentic system would fail or become stuck on larger codebases with limited observability making root-cause diagnosis difficult.

First attempt

The original 'Single, Do-Everything Agent' design produced noisy, low-value review comments and false positives, and the serverless architecture caused intermittent timeouts that the system could not handle gracefully, with no async queuing layer to recover from failures.

Workflow diagram · grounded in source

Pull Request triggers agents

trigger

“filling the gap between user interactions (Pull Requests) and Agent runs”

Planner assesses changes

ai_action

“Planner: Quickly assesses changes and identifies necessary checks.”

Specialized agents run in parallel

ai_action

“Security Agent: Detects vulnerabilities such as injection or insecure authentication. Duplication Agent: Flags repeated or copied code.”

Filtering Agent validates findings

validation

“Filtering Agent: Designed to filter for false positives and confirm issues found.”

Review output to Pull Request

output

“quickly identify bugs, suggest fixes, and generate diagrams of architectural changes, freeing developers from low-level review tasks”

Long-term memory learns codebase patterns

feedback_loop

“cubic's AI Agents seamlessly integrate with GitHub to build a long-term memory that learns codebase patterns over time”

Reported outcome

Transitioning to a multi-agent system orchestrated by Inngest resulted in a 51% reduction in false positives and enabled teams to merge Pull Requests 4x faster, while replacing manual log tagging with fine-grained, searchable traces.

Reported metrics

False positive reduction51%

Pull Request merge speed4x faster

Reported stack

InngestGitHub

Source

https://www.inngest.com/customers/cubic

Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Transitioning to a multi-agent system orchestrated by Inngest resulted in a 51% reduction in false positives and enabled teams to merge Pull Requests 4x faster, while replacing manual log tagging with fine-grained, se…

What tools did this team use?

Inngest, GitHub.

What results were reported?

False positive reduction: 51%; Pull Request merge speed: 4x faster (source-reported, not independently verified).

What failed first in this deployment?

How is this quality assurance AI workflow structured?

Pull Request triggers agents → Planner assesses changes → Specialized agents run in parallel → Filtering Agent validates findings → Review output to Pull Request → Long-term memory learns codebase patterns.