European software company Cube sees 400% ROI with GitLab

When Cube rolled out AI-assisted development with GitLab in 2024, the value showed up quickly. Not just in faster code writing, but in how teams used AI to navigate the context of each project: codebase, issues, merge requests, pipeline history. That kind of context-aware assistance only works when everything already lives in one place.

Adoption spread fast, and AI features became part of every DevSecOps team's daily workflow at Cube. But the team didn't stop at code assistance. They started building custom agents on GitLab Duo Agent Platform, each one targeted at a specific bottleneck across the SDLC:

• Refinement agent: Reviews incoming tickets to determine what work is actually required, surfacing scope, dependencies, and open questions before a ticket reaches a developer.

• Context sub-agents: A set of supporting agents that figure out which path to take on a task, gathering context across the codebase, related issues, and prior work so developers start the dev phase with the full picture.

• Design reuse agent: Helps the design team check whether components or patterns have already been designed in earlier projects, cutting duplicate work and keeping design consistent across clients.

• Maintenance agent: Scans every project for major and minor upgrades that need to be applied, categorizes the changes by risk, and flags what needs attention.

Every one of these agents is built within Duo Agent Platform and managed inside GitLab, alongside the rest of the development work. Cube also exposes GitLab context through the Model Context Protocol (MCP) server, which lets external AI tools like Anthropic's Claude reach the same project data for tasks that originate outside the platform, such as project intake or cross-project analysis. With multiple development teams running dozens of client projects in parallel, consistency matters. The same agents and the same context layer work across every project, which avoids the fragmentation that happens when teams use AI tools that each store their own context in isolated systems.

"GitLab Duo Agent Platform sits right where the work happens," says Booijink. "Our agents have native access to every issue, every merge request, every pipeline, and that context is what makes them useful. When we expose that same context to external tools through MCP, they plug into the same foundation. GitLab is the platform underneath all of it." Mans Booijink, Operations Manager, Cube.

When Cube upgraded to Ultimate, automated security scanning was a key driver. SAST, DAST, dependency scanning, and secret detection were rolled out across all projects immediately. That decision helped Cube achieve both ISO 27001 and NEN 7510 certification. Every new piece of code runs through automated scanners with approved rules, ensuring vulnerabilities are handled consistently regardless of who writes the code. As custom agents gained access to more project data, Cube also developed internal AI governance guidelines covering security risks across all tools and integrations.

GitLab also functions as the governance layer for AI. Whether an agent runs natively on Duo Agent Platform or an external tool connects through MCP, GitLab is where access is scoped, actions are logged, and context is controlled. Cube decides what data an agent can see, which repositories it can touch, and what it can change. Every step shows up in the audit trail next to the rest of the delivery timeline. That makes AI workflows subject to the same governance as any other part of the SDLC, which is what makes them defensible under ISO 27001 and NEN 7510.

"It's easier to develop secure software without losing any speed. We have automatic scanners that go over all new code, and approved rules that ensure every vulnerability is handled the right way. That matters even more now that custom agents work on the same codebase. They operate inside the same guardrails, not around them." Mans Booijink, Operations Manager, Cube.

Getting custom agents into production takes more than turning on Duo Agent Platform. Cube built the foundations to make adoption stick: internal AI governance guidelines, role-based access patterns for agents, and a skills infrastructure that captures how each team uses AI in their workflows.

Having everything in one platform makes adoption easier. New Duo Agent Platform capabilities and MCP integrations land in the same environment where teams already work. No new tool to learn, no new login, no context switching.

GitLab stays at the heart of the setup, even when external AI tools come into play. The refinement, context, design reuse, and maintenance agents all run on Duo Agent Platform with native access to GitLab data. External tools like Claude plug in through MCP and read from the same data layer. The model behind an agent can change as better options emerge, but the platform and the data underneath stay the same, so swapping a model is a configuration change, not a rebuild.

Running on GitLab Ultimate with Duo Agent Platform translates into measurable outcomes:

• 50% faster project setup. Templates, pipelines, and issue structures are prepared with help from custom agents. Project kickoff went from a full week to a couple of days.

• Multiple production deploys per day, a x5 increase. Automated scanning, AI-assisted review, and CI/CD in one platform move teams into the DORA elite performer category.

• 400% ROI on GitLab Ultimate with Duo Agent Platform. Hours saved across code writing, reviews, security, and project setup add up to four times the cost of the investment.

As an official GitLab Channel Partner, Cube brings a unique perspective. They don't just use GitLab. They implement and consult on it for other organizations. Every feature they adopt internally becomes knowledge they can share with clients.

From CI/CD pipeline setup to AI-assisted development rollout and now custom agent development on Duo Agent Platform, Cube's hands-on experience translates directly into practical guidance for organizations starting their own GitLab journey. Learn more about Cube's GitLab consultancy services.

Cube is expanding its use of Duo Agent Platform and building more custom agents to automate repetitive work across the SDLC so teams can focus on solving problems for clients.

"We chose GitLab in 2018 to consolidate our toolchain. Eight years and several major shifts later, we're still on the same instance, running custom agents our 2018 selves couldn't have imagined. That's not luck, that's what a good platform does: it grows with you instead of holding you back." Booijink.