The Sprint Is Dead — We Just Haven’t Buried It Yet
Earlier this week I came across an AWS session on something called AI-DLC — AI-Driven Development Lifecycle. It had been sitting in my watch list for a while. I finally got around to it.
One data point in particular didn’t let me go. A study by Metr.org ran a controlled experiment with 16 experienced open-source developers working through roughly 250 real issues. Developers using AI tools felt 23% more productive than their baseline. They were, in reality, 19% slower than the control group working without AI.
That gap — between perceived and actual productivity — is, I think, the most honest summary of where most organizations are right now with AI in software development. And it’s the starting point for why what I heard in this talk has stayed with me.
The problem isn’t the tools. It’s the process architecture around them.
AWS identified two dominant failure patterns they’ve observed across more than 100 companies, from startups to Fortune 100:
- AI Managed: Throw complexity at the AI and expect autonomous output. The result is mountains of code nobody fully understands, no trust, no go-live.
- AI Assisted: Keep the old process intact, use AI only for narrow, isolated tasks. The result is that meetings, dependencies, and planning overhead absorb every efficiency gain before it reaches the bottom line.
Both patterns share the same flaw: they graft AI onto a process logic designed for a pre-AI world. The sprint, the backlog, the two-week cadence — these weren’t arbitrary inventions. They were rational responses to a specific constraint: that software development was slow, expensive, and unpredictable. The sprint exists to manage uncertainty in that environment.
What happens to that logic when the constraint disappears?
AI-DLC doesn’t speed up the old process. It replaces the assumptions underneath it.
The methodology’s core ritual is what they call Mob Elaboration: all stakeholders — PMs, developers, QA, ops — in one room for four hours. AI refines intent, generates user stories, everyone validates in real time. Alignment in hours instead of months. What used to take quarters happens in a morning.
The results from early adopters are hard to dismiss as outliers. A global IT consultancy ran a cross-border healthcare project originally scoped for several months. It closed in 20 hours. A FinTech built a new trading app planned for two months. They shipped in 48 hours, with a live release one week later.
These aren’t productivity improvements. They are a different category of outcome entirely.
My view
The two-week sprint isn’t dying because AI writes code faster. It’s dying because the bottleneck is moving. The sprint assumes development is the slowest link in the chain. That assumption is breaking down. When a feature can go from decision to working code in a day, the limiting factor becomes something else entirely: clarity of requirements, speed of decisions, quality of deployment pipelines, governance. The process overhead that Agile was built to minimize now is the bottleneck.
That has implications well beyond engineering teams. Product management, budget cycles, organizational design, how we think about capacity and headcount — all of it was calibrated to a delivery pace that is structurally changing.
I don’t think most organizations have fully registered what that means yet. I’m not sure I have either.
It was a good talk. I’ll be spending more time with this.
The talk was delivered at an AWS developer event. The AI-DLC workflow is available as an open-source steering file set for tools like Amazon Q and Kiro.
Consultant's Digest 
Hinterlasse einen Kommentar