Why Kanban Matters More in the AI Era

Let me say this up front: I’m all in on AI. I use it every day; it’s genuinely changing how I work – and how our teams work.  The ones who ignore it will fall behind. This isn’t a “slow down, AI is dangerous” piece.

But being all in isn’t the same as following blindly. AI is brilliant at the what — it generates, drafts, summarizes, and surfaces work faster than any team could before. What it doesn’t do is own the why: why are we doing this, why now, why this over the rest of the things in the backlog, etc. 

That judgment stays human. And the place where human judgment meets the flow of real work is a Kanban board.

That’s why I think Kanban in the AI era matters more, not less — and why the two aren’t in tension at all. As AI gets better at producing output, the discipline of managing flow, and keeping human judgment inside it, becomes more valuable. Kanban is how you get the most out of AI without losing the plot.

AI gives you the ‘what’. Humans still own the ‘why’. Of course!

Drop a task into an AI tool, and you get output in seconds. That feels like progress. But output isn’t the same as value, and speed at one step isn’t the same as delivery.

A Kanban board exists precisely to make that distinction visible. It shows where work actually is, where it’s stuck, and what’s waiting on a human decision. AI can populate that board faster than ever. It still can’t tell you – at least not yet –  whether the work should move — that’s the exit criteria, the priority call, the “is this actually done?” judgment that people own. Strip the human “why” out of your flow and you don’t get a faster team. You get a faster way to produce things nobody needs.

Is Kanban still relevant with AI? More than ever — because of where the work piles up

The AI Engineering report 2026, by Faros.ai, captured AI’s impact on engineering orgs very well. One of its findings argues that when AI writes code roughly ten times faster, the system doesn’t necessarily get faster. Rather, the bottleneck jumps downstream — to code review, testing, integration, and product validation, the stages that still run at human speed. This was phrased as ‘acceleration whiplash’ in the report. 

Created using Gemini

What you get as a result is a queue. Pull requests pile up at the review gate, either waiting for human eyes or getting waved through without real understanding. Yuval Yeret, an agile coach and AI transformation advisor, points out that this is a system-level flow problem, not a tooling problem — and the answer isn’t more AI. 

It’s the oldest Kanban practice there is: limit work in progress, so speed at one stage doesn’t flood the next. He calls the alternative “AI theater” — maximum activity, minimal impact. (We’ve written before about why AI projects fail; this is one of the quiet reasons.)

The Kanban ideas worth revisiting now — and one worth retiring

Kanban practitioners are already rethinking the craft, independent of AI. José Casal, a long-time Kanban teacher, has been revisiting practices he once taught. His points weren’t about AI at all — but read them in an AI context, and they land even harder.

Take the language. Casal argues we should stop saying “work in progress” and start saying “unfinished work” or “half-done work.” “In progress” sounds like motion; “unfinished” tells the truth — value not yet realized, a commitment not yet kept. 

In an AI era where it’s trivial to start a dozen things at once, that reframing matters. A board full of AI-generated drafts isn’t a board full of progress. It’s a board full of unfinished work.

His reframe of the classic mantra fits the moment too. Instead of ‘stop starting, start finishing,’ Casal prefers ‘be slow to start and fast to finish’ — begin work only when you have the capacity and dependencies to actually complete it. When AI makes starting almost free, that discipline is the whole game.

A few more of his points that apply directly:

• Design columns around the state of the work, not roles. A board that just mirrors an org chart hides the real flow.
• Make waiting visible. In low-flow-efficiency systems, work spends most of its life waiting, not being worked — and AI, by speeding up the “being worked” part, makes the waiting even more glaring.
• Manage load at the system level, not just per column, because in complex knowledge work the bottleneck is dynamic — it moves between testing, product ownership, and dependencies. Which is exactly what “acceleration whiplash” describes.

None of this requires AI to be true. But AI raises the cost of ignoring it.

Your Kanban board is already your AI-adoption blueprint

Here’s the idea I keep coming back to, and it’s one I fully agree with. A well-run board has explicit exit criteria on every column — the rules for what have to be true before work moves to the next stage. Most teams treat those as process hygiene. They’re actually the most practical starting point for AI adoption anyone has.

You don’t need a strategy document, a committee, or a vendor pitch to figure out where AI fits. You need your board. The exit criteria tell you exactly which checks, drafts, and hand-offs are candidates for AI assistance — and, just as importantly, which decisions must stay human. 

The flow metrics tell you whether the AI you added actually improved delivery or just added noise. It’s all right there. (If you’re formalizing this, our guide on when and how to implement Kanban is a good starting point.)

How AI changes flow-based delivery: the board as a guardrail for agents

There’s a story I saw from a developer that captures this perfectly. Every time they asked an AI to add a feature, the agent had no context for what was coming next — so it would ‘optimize’ the architecture in a way that broke the plan for the next three features. More cleanup than time saved.

The issue was these AI coding agents frequently optimize for the immediate feature request using only the context currently loaded into their window, not for the broader roadmap. So, what was the fix?

Their fix wasn’t a better prompt. It was a Kanban board. They gave the AI agents a board where the full roadmap was visible: a planner agent checked each new task against the roadmap and the existing architecture, broke it into steps, and handed it off to a builder agent — with a human approving each step before it moved forward. The board became the shared context and the guardrail, and the human stayed in the loop at every transition.

That’s the pattern. As agentic tools do more of the doing, the board helps you keep them oriented, and how you keep a human on the “why.”

Kanban or Scrum for AI-assisted teams?

In these “Agile is Dead” times, this may no longer be a relevant question.  Yes, it may be worthwhile considering the question: does Kanban work better than  Scrum for AI-assisted teams? 

For genuinely exploratory AI work — model prototyping, tuning, shifting priorities — Kanban’s continuous flow and flow metrics tend to fit better than fixed sprints. 

For productized AI features that need a predictable release rhythm, Scrum still earns its place. It isn’t a rivalry; it’s a fit question, and plenty of teams use both – as Scrumban. The constant across both is flow discipline — that’s the part AI makes non-negotiable.

What this means for delivery leaders

If your teams are drowning in AI output but delivery hasn’t sped up, don’t buy another tool. Tighten your work-in-progress (and maybe call it “unfinished work”!). Make your exit criteria explicit. Keep a human gate on the decisions that matter. The scarce resource in the AI era isn’t output — it’s judgment and flow, and Kanban is how you protect both.

This is also, quietly, where tooling earns its keep: a board built for flow — with real WIP Limit control and flow metrics like cycle time, flow efficiency and throughput — is what keeps AI-accelerated work honest. 

It’s why we think of Nimble as the delivery intelligence layer for human and agentic teams, not just a place to park tasks. (For choosing tools without the hype, see how to evaluate AI in PM tools.)

The teams that win the AI era won’t be the ones generating the most. They’ll be the ones who kept flow honest and kept humans deciding the why.

FAQ

Is Kanban still relevant with AI? 

More than ever. AI speeds up how quickly work gets produced, which pushes the bottleneck to human review and validation. Kanban’s core practices — visualizing flow and limiting work in progress — are what stop that speed from piling up into downstream chaos. Another factor – as teams deliver more and more often, 2-3 week sprints may no longer make sense, if they have their DevOps pipelines and automation in place. Kanban’s focus on continuous delivery makes it even more relevant in the AI-assisted era.

How does AI change Kanban / flow-based delivery? 

AI accelerates the “doing” stages, so waiting and review stages dominate the timeline. That makes flow discipline — WIP limits, explicit exit criteria, and making waiting visible — more important, and it makes the board an ideal place to give AI agents shared context and a human approval gate.

Kanban vs Scrum for AI-assisted teams? 

Kanban tends to suit exploratory, R&D-style AI work with shifting priorities; Scrum suits predictable, productized AI feature delivery. Most teams blend them. The common requirement is disciplined flow.  Most stable teams and organizations do a combination of both types of work at any given time, so our recommendation: combine both and use Scrumban, just like we do!

Working in flow-based delivery and rethinking what AI changes? Sign-up for updates and  follow along!