As companies build AI-augmented development teams and roll out AI agents, the core challenges of improving flow, delivering services and becoming more fit for purpose will remain important goals. Even as development velocity increases, new parts of the business will become bottlenecks, slowing the flow of value to customers.
There is always a bottleneck. And so, as AI eliminates the friction of writing code, the bottleneck doesn’t go away; it simply shifts elsewhere in the value stream. The bottleneck will likely increasingly fall in slower, non-technical parts of organizations. For example, there will be increasing demand for high-velocity decisions on what to build and on understanding the impact of work delivered to customers.
Instead of just optimizing team-level software delivery, Kanban has been one of the few methods that help leaders manage and mature their business. Kanban’s focus on customer fitness, end-to-end flow, and leadership makes it a useful tool for organizations seeking to realize the full value of AI.
Deciding What to Build
When AI lets you build in minutes instead of weeks, dramatically lowering development costs and increasing capacity, one of the most important decisions becomes choosing what to build.
While some frameworks focus on moving items from a product backlog to development done, Kanban is focused on improving the end-to-end flow of work, from idea to delivery. The practice of Upstream Kanban provides organizations with the tools needed to manage product discovery activities.
Upstream Kanban, also commonly referred to as “Discovery Kanban” or “Customer Kanban,” is concerned with activities including:
- Generating ideas
- Managing a pool of ideas
- Testing those ideas
- Validating the business case for ideas
- Idea triage
Originally, the job of Upstream Kanban was to ensure that software development teams would always have just enough good options for what to build and deliver. It would be wasteful not to have anything for teams to work on, but also to generate hundreds of options that teams couldn’t actually work on.
But to support the AI-augmented development pipelines of the future, companies will need to increase collaboration and integration between their upstream teams and the rest of the value stream. Instead of handing development teams annual product plans, the process will need to be more integrated, collaborative, frequent, and agile.
Manage the Work, Not the People
Manage the work, let people self-organize around it.
In an AI-driven world, the relationship between developer and work changes. For every developer, an organization could have a hundred AI agents, all autonomously creating tasks and delegating them to other agents to develop, review, and deploy.
Because Kanban is focused on the flow of the work, it’s unaffected by who’s doing the work, AI or human. Instead, it’s more concerned with improving the health of the system, using flow-based measures such as:
- Customer Lead Time: How long does it take from the moment a customer expresses a need to the moment value is delivered?
- Flow Efficiency: How much of its total lead time does work spend waiting?
One way Kanban and these metrics help is by giving organizations new and better perspectives on work and process improvement, including a shift from increasing individual effort to optimizing for flow.
Evolutionary Policies vs. AI Micromanagement
Evolve your management policies to improve customer & business outcomes
As AI enters the workplace, there will be a strong temptation for organizations to enforce control-based project management practices, including:
- Using AI as an excuse to exert more control over teams.
- Designing strict processes on how teams should build with AI, pushing teams into a new form of big batch, quality-gated waterfall.
- Using Gen AI to create new forms of control that were not possible before
Kanban’s Principles and practices offer companies a way to develop AI-augmented ways of working that avoid micromanagement. Through a combination of Kanban’s practices of visualizing work, managing flow, and making explicit policies, teams can co-create healthier, customer-driven work systems.
Kanban’s Principles are also tool- and process-agnostic; they don’t promote or require a company to adopt any particular way of working. Because AI is highly volatile with models, vendors, regulations, techniques and new development practices changing monthly, the last thing you want to do is prematurely lock-in and create a highly constrained process.
A better strategy is to lean into Kanban’s practice of evolutionary change, letting teams develop processes through cycles of collaborative, continuous improvement.
Conclusion
Because of the impact of AI, every part of a business, including the non-software-development parts of a company’s value stream, will need better management. Investing in 100x engineers won’t make much difference if their impact is limited by organizational bottlenecks either upstream or downstream of development.
“The Kanban Method is not a software development life-cycle process. Nor is it a project management process. It’s something to help you manage your business better.” — David Anderson
The Theory of Constraints teaches us that optimizing anything other than the bottleneck does not improve the system’s performance. Companies that use AI to improve their development activities will just see the bottlenecks shift elsewhere. As new bottlenecks light up across the value stream, methods or frameworks that focus on only managing development won’t suffice. We need to adopt ways of working and systems designed to manage and improve the performance of the whole business.
