“Three days.”

A developer throws out the number confidently in a planning meeting, the team lead writes it down, and the roadmap updates, so on paper it looks managed, but in reality, that number is fiction. 

The Illusion of Accuracy

The developer isn’t lying, they’re relying on muscle memory, as their brain is estimating based on how long a similar feature took in the past, back when execution followed a more predictable, human pace, but that world no longer exists. 

From Linear to Exponential Uncertainty

In today’s environment, the same task can look completely different. With the right context, the model can complete it in under an hour, while without it, the task can stall for days.

The range of outcomes has fundamentally changed, and estimation error is no longer linear, it’s exponential.

Forcing Human Models on Machine Systems

Despite this shift, most organizations are still trying to apply the same frameworks, using story points, sprint commitments, and fixed timelines, effectively forcing human-based estimation models onto systems that operate at machine speed, and it doesn’t work. 

When Planning Stops Reflecting Reality

An architect I spoke with recently put it this way: “Forget sprints. The whole concept is broken. You don’t actually know how long things take anymore. You prioritize, you push work in, and it gets done.”

This isn’t just frustration, it’s a reflection of a deeper mismatch between how work is planned and how it actually executes.

The Cost of False Certainty

Sprints provide a sense of control, with a defined scope, a timeline, and a commitment. But in a world of exponential variability, that certainty becomes misleading.

Teams either overcommit and miss deadlines, or undercommit and underutilize their capacity, and either way, planning drifts away from reality.

Toward a New Execution Model

As the number of parallel tasks increases and the variability of execution expands, organizations need a different approach. 

Less focus on predicting exact durations.
More focus on:

• Clear prioritization

• Continuous flow of work

• Dynamic adjustment based on real progress

• Systems that can adapt in real time

This is closer to a flow-based model than a time-boxed one.

Letting Go of the Sprint Illusion

This is not an easy shift. Sprints offer comfort, providing a structured way to reason about progress, and letting go of that structure means letting go of perceived control.

But holding on to it creates a bigger problem, a growing gap between planning and execution.

Instead of asking how long things will take, start asking how your system adapts when reality changes, because that is the new foundation of execution in the age of AI.