As AI tools take on more of the work of writing, testing and refactoring code, engineering teams are running into new constraints. That is the central message from Claude Code in a blog post describing how its own organization has adapted to agentic coding becoming the default workflow.
In the post, Fiona Fung, director of engineering for Claude Code and Claude Cowork, says the traditional assumptions behind software planning and delivery were built around a world where human coding time was scarce. In that environment, teams relied on long planning cycles and process-heavy coordination to make the most of limited engineering bandwidth. Now, she argues, the bottlenecks have shifted.
Instead of code creation slowing teams down, the harder problems are verification, code review and security. Claude Code says its engineers can produce code quickly with AI assistance, but that speed creates fresh questions about correctness, maintainability and oversight.
One of the clearest changes described in the post is planning. Fung says the team has moved away from heavily front-loaded roadmaps and toward more just-in-time planning. The reasoning is straightforward. If code can be produced and revised much faster than before, long-range plans are more likely to become outdated before work is finished.
The post suggests that this has changed the role of planning from detailed prediction to more flexible coordination. Rather than locking in a lengthy roadmap and following it through unchanged, the team now expects priorities to evolve alongside the work itself.
Claude Code frames this as a broader organizational shift, not just a tooling change. The company says the same processes that once helped manage expensive coding time can become less useful when that assumption no longer holds.
The blog also highlights code review as a key pressure point. With AI generating more code more quickly, human reviewers may struggle to keep pace if they rely on older review habits. Fung says this is one of the questions she is most often asked by other engineering leaders.
The post does not offer a single universal solution, but it makes clear that review practices have to evolve if teams want to preserve quality. That includes thinking carefully about how much can be delegated to AI and where humans still need to stay closely involved.
Security is another area that the company says demands more attention in an AI-native workflow. Faster code production does not remove the need to inspect what is being built. Instead, it increases the importance of validation.
Fung places the shift in historical context, comparing it with earlier changes in software distribution. She recalls work in the early 2000s, when software shipped on physical media and release timing was tied to manufacturing. Later, software moved online and updates became continuous. Now, she says, the core change is centered on the time and people required to write software.
Claude Code presents its experience as an example of what an AI-native engineering organization may look like in practice. The company’s point is not that process no longer matters, but that process needs to match the new reality of AI-assisted development.
For engineering leaders watching the spread of agentic coding tools, the post offers a glimpse of a management challenge that is likely to become more common. As coding gets faster, the work around coding may become the part that needs the most redesign.