Is AI breaking our development cycles?

One of the things I keep thinking about lately is whether AI is quietly breaking some of the assumptions modern software development has been built around for decades.

Not because Scrum, Agile, or iterative development was wrong. But because many of those ideas were created in a world with fundamentally different constraints.

For years, one of the core ideas in software development has been the importance of short feedback loops: Two-week sprints, fast iteration, continuous learning, and close collaboration with the business. The logic was sound: If we work too long without feedback, we increase the risk of building the wrong thing.

But what happens when the amount of work that can be produced inside those feedback loops suddenly changes dramatically?

I already see developers using tools like Claude Code and AI agents becoming two or three times more productive in certain contexts. Work that previously took weeks can suddenly be done in days.

And that creates an interesting question: If a two-week sprint now contains what previously represented four, eight or even more weeks of output, are we still operating with short feedback loops? Or have we unintentionally recreated long-cycle development again, just hidden inside the same calendar structure?

I believe this is only the beginning. Right now, AI often feels similar to the early days of computers. At first, computers mostly replaced manual bookkeeping and typewriters. Existing processes remained largely intact. Things simply became faster and more efficient.

But the truly transformative shift came later with computers connected via networks, the internet, cloud computing, and SaaS platforms. All enabled entirely new operating models. My bet is that AI will follow a similar trajectory.

Today, many developers still write code directly, just faster and with more assistance, equivalent to the introduction of electronic bookkeeping and word processing. But agentic coding is already emerging, and tools like Claude Code give us a glimpse of something much bigger, even if I do not think it has become mainstream yet.

I believe developers will increasingly orchestrate agents instead of writing most of the code themselves. If you already know how the code should be written, there is a growing argument that you may no longer need to write every line manually. You can instead guide agents that generate code in your preferred style, architecture, and quality standards.

At that point, the role starts shifting. The developer becomes less of a pure producer and more of an orchestrator, reviewer, architect, and decision-maker.

And if productivity increases not by 2x or 3x, but by 10x or 20x over time, then the implications become much bigger than individual productivity gains. Because eventually, software development itself may stop being the bottleneck.

Everything around it becomes the bottleneck instead: Pull request reviews, governance processes, manual testing, approvals, meetings, deployment pipelines, decision-making structures, organizational coordination, etc.

Many organizations are currently optimized for a world where writing software is expensive and relatively slow. What happens when software generation becomes almost instantaneous, but organizational decision-making still moves at human speed? The entire system starts looking different.

And that raises another interesting question: What happens to Scrum and fixed sprint cycles in a world where production speed accelerates dramatically?

Now, to be fair, I know many organizations have already moved toward shorter cycles, continuous delivery models, Kanban-inspired flow systems, and more flexible ways of working. Some have already moved away from classic Scrum entirely.

But I think the implications here go much deeper than whether a team runs two-week sprints or not. Because this is not just about software development teams. This potentially changes the speed of the entire system around product development: Customer feedback, marketing, experimentation, product discovery, deployment, operations, governance, decision-making, etc.

Every process connected to building, launching, and evolving software products. And if software creation accelerates dramatically, then all surrounding organizational systems suddenly come under pressure to evolve as well.

This is part of why I increasingly believe the future may move further toward instant deployment, flow-based systems, Kanban-inspired thinking, and highly automated operational models. Not because Scrum was wrong, but because the assumptions underneath Scrum are changing.

If one or two people with AI agents can suddenly achieve what previously required large teams, then team dynamics themselves start changing. Instead of simply evolving existing ways of working, we might have to invent entirely new ones.

• What does leadership look like in that world?
• What does coordination look like?
• What happens to roles?
• What happens to organizational design?

And to address a specific example: What happens to retrospectives?

Will AI agents eventually analyze delivery patterns, incidents, collaboration bottlenecks, deployment failures, and customer feedback continuously in real time? Will improvement loops themselves become automated? Will retrospectives evolve from scheduled meetings into always-on learning systems?

And if teams increasingly become humans orchestrating collections of agents, is there even a need for traditional Scrum Master capabilities in the same way we know them today? Or does orchestration itself become something fundamentally different?

I genuinely do not know. But I am increasingly convinced that we are looking at something much bigger than productivity gains. We are looking at a shift in how work is designed, coordinated, led, and improved.

Regardless what the future holds, I am very excited to be part of it and help shape how software, teams, and organizations work in the coming decade.

What do you think?