AI helps teams produce software faster, but review and approval processes can delay its release. A Kanban workflow makes these bottlenecks easier to manage.

Summary

• AI-assisted development is speeding up software production, but review, approval and governance processes can still slow delivery.
• Kanban helps AI Build teams see where work is waiting, which stages are overloaded and what is preventing items from reaching production.
• Work-in-progress limits and visible governance checkpoints help teams focus on completing valuable work rather than starting more than they can finish.

By Melani French

---

This is the final article in a three-part series about the Flow Master role in AI Build teams. The first explained why AI-enabled delivery requires more than traditional Scrum facilitation. The second examined the skills Flow Masters need to manage specifications, validation, governance, priorities and decisions.

All these responsibilities require the Flow Master to understand how work progresses from an idea to production, where it slows down and why.

Does a sprint-based framework still give AI Build teams the clearest view of their work? For some teams, Kanban may be a better fit.

Scrum remains useful in the right environment. Its regular planning cycles, meetings and sprint boundaries give teams structure and focus. However, AI is changing the pace of work across the software development process, especially within AI Spec Driven Development teams.

Code, documentation and technical options can now be generated quickly. Code reviews, architectural assessments, security checks, quality assurance, compliance reviews and business approvals still rely on human expertise and accountability. These stages often determine when software reaches production. A team may complete development ahead of schedule and then spend days waiting for an architectural review or approval. A feature may move from idea to implementation quickly, only to be held up by several reviews, governance checks and approval queues.

Sprint completion alone may not reveal these delays. The team can appear productive while completed or partially completed work builds up between stages.

Review and approval are slowing delivery

Software teams have traditionally been limited by how much work they could produce. Could developers write enough code in a short sprint? Could testing keep pace with development? Could the team meet the demand for new features in time? AI gives teams the ability to produce more work in less time. Reviewing and approving that work is now becoming the greater challenge for many AI Build teams™.

With today’s technology, AI generated code still requires human review. Technical recommendations need to be checked. Security risks must be assessed, and releases still require approval.

As teams produce more code, review queues grow. Security teams become overloaded. A small group of architects may need to review and make decisions for several projects at once. Business stakeholders are asked to review and approve more changes in less time.

The team may be working efficiently, but delays are happening elsewhere. Teams need a way to see exactly where work is being held up.

Why Kanban works well for AI delivery

Kanban organises work around movement and flow, rather than fixed time periods. It shows how an item progresses through each stage, how long it remains there and what prevents it from moving forward.

This helps AI Build teams see the pressure that faster software production places on review, testing and approval processes.

Kanban shows where work is waiting, which review stages are overloaded and where approvals regularly cause delays. These problems can occur whether a team uses Scrum, Kanban or a combination of the two. Kanban makes them easier to identify and manage.

This information may be more useful than simply measuring how much work the team completed during a sprint. A team may be able to produce more than the rest of the organisation can review, validate and approve.

Showing every stage of the work

A Kanban board can be designed around the path software follows through an organisation.

That path usually includes much more than development. It may cover specification review, code review, quality assurance, security checks, governance approval and release preparation.

A workflow could look like this:

Backlog → Prioritised → Intent → In progress → Code review → Sanity check → Quality check → Security and Governance → Ready for release

Each stage represents a specific checkpoint. Review, validation and governance appear on the board instead of being grouped under broad categories such as “in progress” or “done”.

The team can identify where work is building up before the delay affects a delivery commitment. Repeated problems also become easier to spot. If work regularly slows at the same checkpoint, the team knows where changes are needed.

The board show how the organisation actually delivers software. Extra unnecessary columns will complicate matters if they do not represent a specific stage or decision.

Limiting work in progress

Work-in-progress limits may be one of the most useful parts of the Kanban flow for AI Build teams.

These limits stop teams from starting more work (which is tempting) than they can review and complete. Without them, AI-assisted development can produce a steady stream of code and other outputs that creates a backlog of review, testing and governance items.

The amount of activity increases, but less work reaches production and add business and customer value.

If the code-review queue has reached its limit, the team must deal with that backlog (and WIP) before adding more work. This may involve reallocating capacity, resolving an outstanding decision or temporarily reducing new development.

Work-in-progress limits help teams finish what they have started before taking on more. This has been an Agile value for many years know as: ‘’stop starting and start finishing’’.

Making governance part of the process

The second article in this series identified governance as one of the Flow Master’s main responsibilities. A Kanban board can show exactly where governance fits into the software development process.

Security reviews, compliance checks, architectural approvals and business sign-offs can appear as defined stages on the board. They are then treated as part of the work rather than separate activities that happen after development.

For example, a security review becomes a recognised step before release, with a clear limit on how much work the security team can handle and a record of how long each item has been waiting.

This is especially useful in regulated industries, where late governance problems can lead to expensive rework or delayed releases.

Combining Scrum and Kanban

Using Kanban does not mean abandoning Scrum. Many teams benefit from sprint goals, planning sessions, reviews and retrospectives. Scrum teams can also use Kanban practices such as visualising each stage of work, setting work-in-progress limits and measuring how long items take to complete.

Some AI Build teams may adopt Kanban fully. Others may choose a combination of Scrum and Kanban, often called Scrumban. This allows them to continue planning in regular cycles to achieve an epic, while using Kanban to manage the daily movement of work and individual tasks more effectively.

The choice depends on the nature of the work. Scrum may still suit teams with predictable delivery and a preference for short planning cycles. Kanban may be more useful when work reaches review and validation faster than it can be approved. The right approach depends on how the team works.

As AI speeds up software production, teams need to pay more attention to review, validation and governance. The Flow Master helps AI Build Teams™ identify delays and address the problems preventing work from reaching production.

Kanban gives the Flow Master and the wider team within a Spec Driven Development (SDD) team a clear view of where those delays are occurring. This allows the organisation to direct its time and resources to the parts of delivery that need them most.

AI writes software. Specifications guide AI. Flow delivers value. Experts ensure success.

To learn more about DVT’s AI Build teams, visit https://www.dvtsoftware.com/services/artificial-intelligence/ai-build-teams.