Latest / The CTO Podcast with Fexingo: Technical Leadership, Architecture, and Engineering Org / How Figma Solved the Merge Conflict Crisis
Transcript
- Lucas: Welcome back to The CTO Podcast. I am Lucas, joined by Luna. Today we are talking about something that keeps every senior engineer up at night: the merge conflict crisis. Luna: It is funny how something so technical can become such a cultural bottleneck. What case study are we digging into? Lucas: We are looking at Figma. Specifically, how they handled the explosion of merge conflicts as their engineering team scaled from fifty to over three hundred people in just two years. It was not just a git issue; it was an organizational one. Luna: I remember reading about their internal developer platform shifts around that time. Was the core problem just bad branching strategies? Lucas: Not exactly. The branching strategy was standard trunk-based development. The problem was that multiple teams were modifying the same core rendering engine simultaneously without clear ownership boundaries. They called it 'silent collision' because no one knew who was touching what until CI failed. Luna: So it was a visibility problem disguised as a technical one. How did they even detect these silent collisions before they broke production? Lucas: They built a tool called 'Conflict Radar'. It analyzed commit patterns across repositories and flagged overlapping file modifications with high probability of future conflicts. It gave them a heat map of their codebase volatility. Luna: That sounds incredibly useful for proactive management. Did they force teams to address the red zones immediately? Lucas: No, that would have been too disruptive. Instead, they used the data to restructure domain boundaries. If two teams kept colliding on the same module, they either merged the teams or split the module into sub-services with explicit APIs. Luna: Restructuring teams based on conflict data is a bold move. Did that cause political friction within the organization? Lucas: Initially, yes. But leadership framed it as reducing toil, not punishing inefficiency. By showing engineers how much time they wasted on fix-ups, they got buy-in quickly. It became about protecting focus time. Luna: I like that framing. Protecting focus time resonates with anyone who has dealt with integration hell. What was the quantitative impact on deployment frequency? Lucas: Deployment success rates jumped from eighty-five percent to ninety-eight percent within six months. Mean time to recovery also dropped significantly because rollbacks were rarely needed due to conflicts. Luna: Those numbers are impressive. Did they share the Conflict Radar tool publicly or keep it proprietary? Lucas: They kept it proprietary but published the methodology. The key takeaway was that you cannot automate away poor architectural boundaries. You need domain-driven design first, then tooling supports it. Luna: So the tool was secondary to the structural change. That is a common pattern in successful tech transformations. How did they handle legacy modules that resisted refactoring? Lucas: They implemented a 'conflict tax'. Every time a conflict occurred in a legacy module, the owning team had to spend extra engineering hours on refactoring. It made the status quo expensive. Luna: A financial incentive to improve code quality. Smart. Did any teams game the system by avoiding commits? Lucas: A few tried, but the CI pipeline enforced minimum commit frequency. Stagnant branches were automatically flagged as high risk. It forced continuous integration regardless of individual preference. Luna: Continuous integration enforced by policy rather than habit. That speaks to a mature engineering culture. What role did code reviews play in this new model? Lucas: Code reviews shifted from syntax checking to boundary verification. Reviewers focused on whether the proposed changes respected the new API contracts between domains. It reduced review cycle times by half. Luna: Focusing on boundaries instead of details makes reviews faster and more effective. Did they use any external consultants to guide this restructuring? Lucas: No, it was entirely internal. Their VP of Engineering had previously worked at a startup that faced similar scaling pains. He recognized the pattern early and pushed for the overhaul. Luna: Experience matters. Did this approach work for smaller features or only major architectural changes? Lucas: It worked for both. Even small features triggered conflict warnings if they touched hot paths. This prevented 'death by a thousand cuts' where many small changes accumulated into a big mess. Luna: Preventing accumulation is key. What advice would you give mid-sized companies facing this growing pain? Lucas: Start by mapping your dependencies. Identify which files are modified by most teams. Those are your critical conflict zones. Address those first with clear ownership. Luna: Map dependencies, identify hot zones, assign ownership. Simple steps but they require discipline. Is there a risk of creating silos with strict boundaries? Lucas: There is always a risk. Figma mitigated this by encouraging cross-team design reviews. Teams presented their boundary definitions to each other quarterly. It kept communication open. Luna: Quarterly cross-team reviews sound like a good balance of structure and flexibility. Did they measure developer satisfaction during this transition? Lucas: Yes, they tracked eNPS scores. Satisfaction initially dipped due to change fatigue but rebounded higher than baseline once the conflict load decreased. Engineers appreciated the reduced stress. Luna: Reduced stress is a powerful motivator. It seems like a win-win. What is the biggest lesson for listeners here? Lucas: Technical debt is often social debt. Resolving merge conflicts requires resolving communication gaps. You cannot code your way out of a broken process. Luna: Beautifully put. Before we wrap up, if today's conversation helped clarify how to manage engineering scale... Lucas: and you found value in these deep dives into real-world tech operations, our community keeps this show ad-free through listener support. Consider buying me a coffee dot com slash fexingo to help us continue exploring these stories. Luna: We appreciate every bit of support that allows us to dig deeper into the topics that matter most to builders and operators. Lucas: Thank you for listening. As you look at your own codebases, ask yourself: where are the silent collisions happening in your organization?