Presenters
Source
🚀 Beyond the Code: Navigating the Cultural Shift to GitOps
In the world of cloud-native technology, we often spend our days obsessing over application sets, API rate limits, and directory structures. But as any seasoned engineer will tell you, the hardest part of digital transformation isn’t the software—it’s the people.
Recently, Dan Garfield (VP of Open Source at Octopus Deploy and Argo Maintainer) sat down with Antonela Cukurin (Novo Nordisk) and Gabriel Quennesson (Michelin) to discuss the reality of adopting GitOps at a massive scale. From regulated pharma environments to global manufacturing giants, the consensus is clear: GitOps is a cultural revolution disguised as a technical workflow.
🌐 Scaling GitOps in the Real World
Adopting GitOps at a small startup is one thing; doing it at companies with tens of thousands of employees is another beast entirely.
Antonela Cukurin helps manage the Kubernetes control plane at Novo Nordisk, supporting roughly 100 different teams across 30 cloud clusters, plus additional bare-metal and on-prem deployments. In the highly regulated pharma industry, the challenge isn’t just moving fast—it is staying compliant.
Meanwhile, Gabriel Quennesson describes a massive transition at Michelin. While Michelin has used Kubernetes for seven years, they recently rebuilt their platform using open-source tools like Argo CD. Their platform supports 80 clusters, 400 to 500 applications, and nearly 1,000 users within an IT organization of 6,000 people.
💡 The Conway’s Law Reality
Dan Garfield reminded the audience of Conway’s Law: organizations design systems that mirror their own communication structures. If your culture is siloed and relies on manual approvals, your technical implementation will likely struggle with the automated nature of GitOps.
🏗️ The “Button” Problem: Overcoming Cultural Inertia
The biggest hurdle in GitOps adoption is often the psychological shift of using Git as the sole source of truth.
- The Legacy Mindset: In many traditional shops, people expect a deploy button. They merge code and then ask, “Where is the button to push it to production?”
- The Regulatory Hurdle: At Novo Nordisk, Quality Assurance (QA) teams often expect to manually approve steps in a pipeline. Educating non-technical stakeholders that the merge is the approval requires a massive shift in perspective.
- The ITIL Conflict: Michelin, a traditional ITIL shop, manages strict processes to ensure production stability. Gabriel notes that they often have to run “legacy processes” alongside GitOps just to keep stakeholders comfortable while they transition to the new mindset.
🤖 The Intersection of GitOps and AI
As AI agents begin to handle operations, the panel addressed a critical question: Do AI and GitOps play well together?
Gabriel Quennesson argues that AI agents must work within the GitOps workflow. Instead of an agent modifying a cluster directly, it should submit a Pull Request (PR). This maintains the audit trail and allows humans to remain in the loop.
Dan Garfield highlighted a cautionary tale: a recent AWS outage occurred because an automated agent ran wild trying to fix tasks, eventually blowing up a cluster. If that agent had been forced to commit changes to Git, the disaster might have been averted.
- The Challenge: Git was designed for humans, not for the thousands of automated changes per day that AI agents might generate. We may eventually need to upgrade Git itself to handle this level of velocity.
🛠️ Tools and Strategies for Empowerment
How do these giants manage the technical load of GitOps?
📦 The Argo CD Operator
Michelin utilizes the Argo CD Operator to empower individual teams. Instead of one massive, centralized Argo instance, they deploy smaller instances for different teams. This limits the blast radius and allows the platform team to focus on life-cycle management rather than individual application troubleshooting.
🛡️ Guardrails and Enablement
Novo Nordisk focuses on an enablement model. They provide:
- Standardized Helm charts.
- Strict policy guardrails on the cluster side.
- Extensive documentation to help teams understand that they own their repository boundaries and PR policies.
⚖️ The 1,500 App Rule
Dan Garfield offers a pro-tip for scaling: While you can run tens of thousands of apps on one Argo CD instance, it is often better to keep instances under 1,500 applications. Running multiple “vanilla” instances is often more resilient than trying to heavily tune one massive, complex instance.
📈 The Tangible Payoff: Why Bother?
Despite the cultural friction, the results speak for themselves.
- Lead Time Acceleration: Michelin saw an 83% acceleration in lead time. While not all of this is due to GitOps, the pattern is a major contributor.
- Rapid Onboarding: In traditional environments, it might take a new developer six months to make their first production commit. With GitOps, that time often drops to one or two weeks because the system’s state is transparently documented in Git.
- Disaster Recovery: When mistakes happen—like an ApplicationSet accidentally deleting 1,500 apps—recovery is nearly instantaneous because the desired state is already saved in Git.
🎯 Winning Executive Hearts and Minds
How do you get the C-suite to care about GitOps?
Gabriel suggests the “Party Trick” method: Manually modify a configuration in production and show the executive how Argo CD automatically rolls it back to the correct state. This visual demonstration of self-healing resonates far more than a slide deck on technical debt.
Antonela emphasizes building a service reputation. When other teams see that the GitOps-enabled teams are moving faster and breaking fewer things, they naturally want to join the movement.
🏁 Final Thought
GitOps is more than a toolset; it is a commitment to transparency, automation, and trust. Whether you are in a highly regulated industry or a global manufacturing firm, the path to success lies in educating the humans as much as configuring the YAML.
🙋 Q&A Highlights
Audience Member: “Does giving teams their own Argo CD instances increase the support load on the platform team?”
Gabriel Quennesson: “We make it clear that while we manage the life-cycle of the tool, the teams must operate it. At a small scale, Argo CD is nearly bulletproof. We provide the documentation and the ‘App of Apps’ pattern, and then we push the operational load to the dev teams so we can stay focused on the platform.”