By now, throughout 2025, we’ve all noticed that software engineering and the accompanying development lifecycle have dramatically changed. Just two years ago, the longest part of the SDLC was likely the planning and coding of new features, bug fixes, etc. But today, with the adoption of AI coding agents, it’s become clear that writing code is no longer the bottleneck.

In fact, creating code with the help of AI agents has accelerated software development so much that the new bottlenecks are the processes that happen after coding. Source code hosting, code review, integration tests, builds, and deployments are all logjams that are now holding up the increased velocity at which we can produce code.

The bottleneck has shifted from writing code to integrating code.

We are frantically trying to bend our existing tools and platforms to meet this new technology we have at our fingertips. Everyone is racing to provide the next integration point for the next model or agent capability. But is anyone taking a step back and asking if bending our existing tools and platforms is the right approach?

Developer platforms like GitHub, GitLab, Bitbucket, and others are all built on a workflow paradigm that is close to 20 years old now. A paradigm that assumes a human collaboration pattern.

The workflow looks a lot like this on any given weekday:

• I, as a developer, write some code
• I commit that code to my source code host
• I open a pull request to have my code reviewed
• Some CI checks run when my pull request is opened
• Another team member reviews my code
• Once the code is reviewed, I merge that code back to the main branch
• Some more CI checks run to validate the integration back to main
• Finally, I deploy to production.

This approach is both human-centric and siloed. Every step in the journey has historically assumed that some human is involved in moving the code from one silo to the next, either explicitly or implicitly.

Code moving off my laptop to my source code host is me running a commit command. The code moving from my branch to my CI environment is me opening a pull request. My co-worker reviewing my pull request is me assigning them to the review.

This is the workflow we as engineers know like the back of our hand. It’s what feels natural.

So much so that we are all investing a lot of time, energy, and money to optimize each one of these steps.

We have coding agents in the IDE or terminal to optimize writing code. We have code review agents to conduct initial reviews for engineers. We have build acceleration services to make the build process faster so that we get feedback quicker. The list goes on.

However, all of these ideas and innovations are point solutions within the larger paradigm and the developer platforms that host it.

The fundamental truth is that the human-centric model and the developer platforms that host it are the cork in the bottle in this new world where a 10-person engineering team could operate like a 300-person engineering team with the help of agents.

We need to think about how we would redefine the software collaboration paradigm with the new tools available to us. How can we remove the bottlenecks that are being introduced now that exponentially more code is being created? How do we continue to ensure that code is safe to merge, deploy, and is well-tested? When should we flag actual humans to review critical issues?

Answering these questions with our old human-centric paradigm just inhibits the velocity that AI promises. We must rethink our paradigms and harness these new tools to extend the benefits of AI-assisted development into the rest of the SDLC. Because right now, we’re trying to force a fundamentally new way of working into tools designed for a different era.

KubeCon + CloudNativeCon North America 2025 is taking place in Atlanta, Georgia, from November 10 to 13. Register now.