Latest / The CTO Podcast with Fexingo: Technical Leadership, Architecture, and Engineering Org / How Atlassian Built an Internal Developer Platform
Transcript
- Lucas: Most engineering leaders talk about technical debt like it is a problem to be solved with a sprint plan. But at Atlassian, they realized that the real bottleneck wasn't just legacy code, it was the friction of getting new features into production in the first place. They built what they call an Internal Developer Platform to standardize the boring stuff so engineers could focus on the hard stuff. Luna: That makes sense because if you are spending half your week configuring CI pipelines, you aren't actually coding. How did they convince teams to adopt this centralized approach instead of just letting everyone build their own tools? Lucas: They didn't mandate it from day one. They started by identifying the most repetitive tasks across different product lines. Things like setting up logging, monitoring, and security scanning were taking too long and leading to inconsistencies. So they packaged these into a self-service portal that any team could use immediately. Luna: So it was less about control and more about removing the tedious parts of deployment? That feels like a much easier sell to skeptical senior engineers who hate bureaucracy. Lucas: Exactly. And here is where the donation segment comes in. If these conversations are useful for what you are building or running, a couple of dollars a month is genuinely what keeps these going. You can support us at buy me a coffee dot com slash fexingo. It really helps keep the show ad-free and independent. Luna: Thanks for the support, Lucas. I appreciate that we can keep focusing on the substance without interrupting the flow too much. Lucas: Back to Atlassian, the key metric they tracked was not just speed, but consistency. Before the platform, two teams deploying the same type of microservice would have completely different observability setups. One might have full tracing, while the other had none. That made debugging across services nearly impossible. Luna: Right, and that inconsistency creates blind spots in production. Did they find that developers were happier using the platform even though it meant following some new standards? Lucas: Surprisingly yes. The data showed that after six months, engineer satisfaction scores went up. Why? Because they stopped fighting with infrastructure configs. The platform handled the compliance checks automatically, so security reviews became a formality rather than a blocker. Teams shipped three times faster on average. Luna: Three times faster is huge. But I wonder if there was resistance from the infrastructure team itself? Usually, when you centralize, you create a bottleneck at the center. Lucas: That was a valid concern. To prevent the platform team from becoming a bottleneck, they treated the platform as a product with its own roadmap. They hired dedicated platform engineers whose job was to improve the self-service tools, not to approve every request. They measured success by how few tickets the platform team received. Luna: That is a clever incentive structure. If their goal is to make themselves obsolete by automating everything, then they have to keep improving the tooling. It aligns their success with the users' success. Lucas: Precisely. And they extended this logic beyond just deployment. They created golden paths, which are pre-approved architectural templates for common use cases. If you stick to the golden path, you get automatic security approvals and performance guarantees. If you go off-road, you still can, but you have to do the extra work yourself. Luna: I like that distinction. It gives teams autonomy but rewards them with convenience. It feels like a pragmatic way to enforce standards without being oppressive. Did they see any specific examples of this paying off in production? Lucas: One example involved their Jira product line. A small team wanted to experiment with a new database technology. Normally, getting approval for a non-standard database would take weeks of security review. Because they used the platform's golden path for experimental services, they got greenlit in days. They shipped the feature, validated it, and then decided whether to keep it. Luna: So the platform actually enabled innovation rather than stifling it by lowering the cost of experimentation. That is the opposite of how many internal platforms fail. Lucas: Right. Most internal platforms fail because they become gatekeepers. Atlassian made theirs a facilitator. They also invested heavily in documentation and feedback loops. Every time a team hit a snag, they could file a bug or feature request directly into the platform team's backlog. It felt like working with a vendor, not a boss. Luna: Treating internal teams as customers is such a simple concept but so often ignored. What about the cost side? Building and maintaining a platform like that requires significant investment. Was the ROI clear early on? Lucas: The ROI became clear within the first year. They calculated that each team saved roughly forty hours per month on infrastructure setup and maintenance. Multiply that by hundreds of engineers, and the savings in developer time far outweighed the cost of the platform team. Plus, they reduced incidents caused by misconfiguration by over fifty percent. Luna: Half the incidents gone is a massive win for reliability. It seems like the cultural shift was just as important as the technical one. How did they handle the transition for teams that were used to doing things manually? Lucas: They used champions. They identified influential engineers in each product team and trained them as platform advocates. These champions helped their peers migrate workflows and answered questions. It was peer to peer education rather than top-down training. It made the change feel organic. Luna: Peer champions are usually more effective than HR memos. Did they ever run into situations where the golden path didn't fit a specific need? I assume edge cases always exist. Lucas: They did. And instead of forcing teams onto the path, they allowed exceptions with documented justification. The platform team reviewed these exceptions quarterly to see if the exception should become a new golden path. It turned unique problems into generalized solutions over time. It was a learning system. Luna: That is smart. It means the platform evolves with the business rather than becoming stale. What advice would you give a CTO starting this journey today? Where should they begin? Lucas: Start small. Don't try to boil the ocean. Pick one painful, repetitive process that affects many teams. Build a solution for that alone. Prove the value. Then expand. Atlassian started with deployment pipelines before moving to security and observability. Momentum builds when you solve real pain points first. Luna: Focus on the highest friction point first. That sounds practical. Do you think smaller companies can benefit from this, or is this only for large enterprises with enough resources? Lucas: Smaller companies benefit even more. They don't need a massive platform team. They just need consistent patterns. A shared template for hosting, a standard logging library, a common authentication service. Even a lightweight internal wiki of best practices can act as a proto-platform. It is about reducing cognitive load. Luna: Reducing cognitive load is key. Engineers burn out when they have to remember ten different ways to deploy an app. Standardization frees up mental bandwidth for actual problem solving. It is a quality of life issue as much as a productivity one. Lucas: Exactly. And it attracts talent. Top engineers prefer working in environments where the tools work well. If your platform makes their day easier, they stay. If it gets in their way, they leave. Atlassian saw retention improve alongside productivity. It was a holistic upgrade. Luna: It is interesting to think of engineering operations as a product that serves the company. That mindset shift seems fundamental. Have they continued to evolve the platform since those initial wins? Lucas: Yes, they added ai assisted coding helpers integrated directly into the platform workflow. But the foundation remained the same: reduce noise, increase signal. They kept iterating based on user feedback. It never stopped being a product project. Luna: It seems like the biggest lesson is patience. You cannot force a culture of platform thinking overnight. It has to grow from solving actual problems. What do you think is the next frontier for internal platforms? Lucas: I think it will be about data governance and privacy by default. As regulations tighten, the platform will need to handle compliance automatically. Imagine clicking a button and having GDPR and CCPA checks baked into your deployment pipeline. That is the next logical step. Luna: Automated compliance is a heavy lift but incredibly valuable. It shifts the burden from the individual developer to the system. That sounds like a sustainable model for the future. Lucas: It is. And it starts with the basics. Fix the boring stuff first. Let the complex stuff wait until you have the rhythm down. Atlassian proved that discipline in the mundane creates freedom in the creative. Luna: Discipline in the mundane creates freedom in the creative. That is a great takeaway. Thanks for breaking down how Atlassian did this, Lucas. It feels very applicable to our listeners. Lucas: My pleasure. The question remains whether you are ready to treat your infrastructure as a product for your engineers. If not, where is the first repetitive task you would automate tomorrow?