A Cumulative Flow Diagram (CFD) exposes bottlenecks in a system by tracking the arrival and departure of work items across stages. CFDs are commonly used by teams using Kanban to reveal bottlenecks in typical development phases like Analysis, Development, and Testing. But as AI agents take over more of our delivery pipeline, how teams use CFDs needs to change.
Before AI, CFDs helped highlight the queues between different activities. Today, AI agents can collapse those activities into a single, automated workflow that don’t require any human input. When a single prompt can trigger the analysis, architecture, code generation, and test execution for a feature that previously took weeks, what use does a CFD still have in that case?
Even when AI enables teams to produce an exploding volume of work at high speed, constraints don’t disappear, they only move. With AI agents handling routine work, people are left operating in a continuous state of highly demanding high-context, non-routine cognitive work, causing a new set of bottlenecks to emerge.
When Execution Is No Longer the Constraint
Agile encourages teams to develop software incrementally, making small changes to reduce complexity and increase the speed of learning. Today, a user can describe a broad product vision to an AI agent, and the agent can autonomously decompose the problem, determine the architecture, create a plan, generate tasks, distribute them to sub-agents, and write the code from a single input.
A CFD won’t be much use if it only tracks AI throughput, because execution throughput isn’t the constraint.
A modern CFD must capture where human judgment and decision-making exist in the workflow:
- Problem framing and discovery
- Evaluating trade-offs
- Creating cross-team alignment
- Validating or reworking AI-generated outputs
- Reviewing product outcomes
The Loss of “Productive Downtime”
In the pre-AI era, even simple tasks required a lot of waiting, which gave developers breathing room. Developers spent time updating Jira, sitting in meetings, responding to emails, or waiting for dependencies. The friction of everyday work created waste, but it also gave people a chance to rest between cognitively intense tasks.
When AI automates routine tasks, that downtime disappears. Work becomes concentrated in high-judgment decision-making and high-context collaboration—requiring deep focus, rapid context switching, aligning with others and balancing trade-offs.
AI agents can work virtually infinitely, but human cognitive capacity is finite. There is a hard ceiling on the number of complex decisions a person can make in a day before speed and quality drop. Organizations building systems for fast flow will have to prioritize optimizing team cognitive load as much as they do how work moves through a value stream.
Modeling the Real Constraints to Fast Flow
For a CFD to remain useful, its scope must encompass the system’s actual constraints. If the bottleneck lies outside the boundaries of your diagram, the chart stops providing insights and becomes unhelpful.
Tracking local optimizations in Development or Testing does nothing for total system lead time if the real bottlenecks sit either upstream (“What should we build?”) or downstream (“Did we build the right thing?”).
The value of a Cumulative Flow Diagram doesn’t come from tracking Analysis, Dev, or Test. Its value comes from helping you visualize flow and spot bottlenecks across the entire value stream. In an AI-accelerated world, keeping CFDs relevant means understanding and tracking where the constraints in your system are. It’s not only about where work waits, it’s also where people still need to approve, align, synthesize, check, validate, review, discuss, collaborate, decide and think.
