Hi Jurgen, you got me thinking on this one. Here is my 2 cents.
The eternal fight between fuzzy and procedural modelling styles existed long before AI. Content, as you pointed out, is a perfect example, with exploratory steps, feedback loops, and so on before entering a more predictable publishing and distribution phase. AI stressed the current visualisation because it's reshaping virtually every product delivery process.
Hoping to keep the exact board representation for a product with a new (still exploratory) process is a losing battle. The friction of using the old board should prompt the building of a new one or maybe adding a different swimlane during the transition phase.
One recurring pattern when dealing with creative, non-linear processes is to stop modelling the flow altogether: it's not linear, and feedback loops won't bring much clarity to the representation. Focusing on the termination condition usually helps, like in "I have no idea what the creation process is, but I can say when one product is finished." But I've seen many teams digging rabbit holes under the assumption of linear processes.
Digital tools don't help. Physical Kanban boards were more flexible and versatile than most digital counterparts. Put another way, a digital Kanban board offers just a subset of the visualisation possibilities of physical ones (I could envision multiple swimlanes after a "Ready for Publishing" state). It often constrains the board modelling towards linear processes.
So, is the AI-driven process reshaping, forcing us to drop board representation? It's clearly forcing us to redesign our boards and maybe even take a "go back to the roots" approach if we need more flexibility. However, saving the board is not the ultimate purpose: a board is just a tool to manage the flow, with enough accuracy to be a good model (like "constraints in the board will reflect in behaviours in the real world"). Still, a pipeline-driven board, no matter how precise, can be useless if the key constraint is a single person, being a distributed bottleneck on many different steps.
In short, I think the Kanban theory still holds, but practice requires revision, especially if we settled on habits, or if we let ourselves be driven by tools.
Hey Alberto, getting someone as smart as you thinking is a compliment in itself. 😁
Yes, to some extent people will try to capture the increase of nonlinearity with creative adaptations of their boards. All good. But I predict, at some point, we're going to give up and tell our AI agents: "Fuck it, you do this."
Yep, trust would be the tipping point. Supervised AI would create a tension between scalable AI actions (ignoring the energy footprint) and less scalable human supervision. The moment someone drops supervision, columns will magically fold.
But I realise there are two levels in the challenge you describe. AI is 1. Redefining new processes prompts a board/flow redesign, and digital boards tend to be stiffer. This problem is annoying and solvable, but it takes time. 2. AI will provide new options at an accelerated pace, making even the adaptation work too ephemeral and potentially pointless. "Fuck it" becomes a tempting strategy, then.
The AI-driven pace of change is the real crisis factor. Now, would I give up flow management because of this? Probably not. However, there would be some anarchy before finding a more fitting representation.
Think Amazon warehouses and Uber ride sharing. There are no visualizations for such complex logistics. AIs don't need them. Only humans do. We only need visualizations for what we have not yet delegated to algorithms.
Wait a minute! You are the same guy who gave me the seven levels of delegation. 😉
And I found myself using this scale (sometimes with variations) to fine-tune the delegation level between humans and computers when designing processes before and after the rise of AI.
So... Isn't it a variation between 7-full delegation and 6-inquire? I fully agree that visualisation is for a human consumer. Yet, even in a fully managed scenario, like the ones you described, there would probably be some level of supervision. Like managing the average waiting time, and reacting to spikes and anomalies. Not a Kanban board, though, it's more food for a crisis unit.
At the same time, I remember the days before Continuous Delivery, when "Deployment Ready" was a column on the board. The moment this step became automated, fast and reliable, the column stopped making sense.
This was a great read. I think I’m on board with the shift you’re describing—flowcharts make way more sense than rigid boards in a nonlinear world.
That said, I’m curious how teams actually operate inside a system like this. Are there tools today that support modular, dynamic workflows beyond just visualizing them? You can use something like Miro to sketch it out, sure, but execution is the hard part. Diagrams don’t do the work.
Maybe that’s your point though: remove the human from workflow management altogether and let the tools handle the flow themselves.
I'm not sure I understood what "non-linear" means here. The fact the steps can appear and disappear or that the work doesn't always move 'left to right' on a kanban board makes it non-linear?
If that is the case, I would suggest that the work is still done in a linear fashion - it's just that what is worked on and when (and by whom) is dynamic. Being able to respond to this dynamism efficiently is definitely beyond the capabilities of typical kanban.
One thing we would do on my teams is account for the feedback loops by simply creating another work item (e.g., I found a bug, the bug is a new Issue with dependencies and effort and needs to be sequenced in) and accounting for the time and schedule impact using my Agile Projector tooling. The fact that it needed a new visualization was not a reason to stop visualizing it, but rather a reason to make re-visualizing easy. We also built metrics to measure this variance, with the goal of minimizing it (when economical) through better planning and ownership in the flow or just naturally accounting for it in the +/- of schedule projections.
For AI, I prefer to ask the question - is this a deterministic process often enough (e.g., once a day) that we are really looking at a rules engine and good tooling vs. a need for AI? It seems that a list of Issues that is updated once a day with size and dependency for each, and a team with customizable skills, preferences, availability and velocity doesn't need AI to create an optimized schedule.
I agree that it doesn’t make sense to visualize complex work in a linear way. Intrinsically it already must be wrong. I don’t think that it is valuable to visualize the exact process steps for such work. I wouldn’t even try. I think abstraction (fuzzy) is a way to avoid going down the rabbit hole and constantly updating boards or flow diagrams. In the end a board is an information radiator. I guess the information needed depends highly on the context.
Hi Jurgen, you got me thinking on this one. Here is my 2 cents.
The eternal fight between fuzzy and procedural modelling styles existed long before AI. Content, as you pointed out, is a perfect example, with exploratory steps, feedback loops, and so on before entering a more predictable publishing and distribution phase. AI stressed the current visualisation because it's reshaping virtually every product delivery process.
Hoping to keep the exact board representation for a product with a new (still exploratory) process is a losing battle. The friction of using the old board should prompt the building of a new one or maybe adding a different swimlane during the transition phase.
One recurring pattern when dealing with creative, non-linear processes is to stop modelling the flow altogether: it's not linear, and feedback loops won't bring much clarity to the representation. Focusing on the termination condition usually helps, like in "I have no idea what the creation process is, but I can say when one product is finished." But I've seen many teams digging rabbit holes under the assumption of linear processes.
Digital tools don't help. Physical Kanban boards were more flexible and versatile than most digital counterparts. Put another way, a digital Kanban board offers just a subset of the visualisation possibilities of physical ones (I could envision multiple swimlanes after a "Ready for Publishing" state). It often constrains the board modelling towards linear processes.
So, is the AI-driven process reshaping, forcing us to drop board representation? It's clearly forcing us to redesign our boards and maybe even take a "go back to the roots" approach if we need more flexibility. However, saving the board is not the ultimate purpose: a board is just a tool to manage the flow, with enough accuracy to be a good model (like "constraints in the board will reflect in behaviours in the real world"). Still, a pipeline-driven board, no matter how precise, can be useless if the key constraint is a single person, being a distributed bottleneck on many different steps.
In short, I think the Kanban theory still holds, but practice requires revision, especially if we settled on habits, or if we let ourselves be driven by tools.
Hey Alberto, getting someone as smart as you thinking is a compliment in itself. 😁
Yes, to some extent people will try to capture the increase of nonlinearity with creative adaptations of their boards. All good. But I predict, at some point, we're going to give up and tell our AI agents: "Fuck it, you do this."
Yep, trust would be the tipping point. Supervised AI would create a tension between scalable AI actions (ignoring the energy footprint) and less scalable human supervision. The moment someone drops supervision, columns will magically fold.
But I realise there are two levels in the challenge you describe. AI is 1. Redefining new processes prompts a board/flow redesign, and digital boards tend to be stiffer. This problem is annoying and solvable, but it takes time. 2. AI will provide new options at an accelerated pace, making even the adaptation work too ephemeral and potentially pointless. "Fuck it" becomes a tempting strategy, then.
The AI-driven pace of change is the real crisis factor. Now, would I give up flow management because of this? Probably not. However, there would be some anarchy before finding a more fitting representation.
(Still thinking) 🤔...
Think Amazon warehouses and Uber ride sharing. There are no visualizations for such complex logistics. AIs don't need them. Only humans do. We only need visualizations for what we have not yet delegated to algorithms.
Wait a minute! You are the same guy who gave me the seven levels of delegation. 😉
And I found myself using this scale (sometimes with variations) to fine-tune the delegation level between humans and computers when designing processes before and after the rise of AI.
So... Isn't it a variation between 7-full delegation and 6-inquire? I fully agree that visualisation is for a human consumer. Yet, even in a fully managed scenario, like the ones you described, there would probably be some level of supervision. Like managing the average waiting time, and reacting to spikes and anomalies. Not a Kanban board, though, it's more food for a crisis unit.
At the same time, I remember the days before Continuous Delivery, when "Deployment Ready" was a column on the board. The moment this step became automated, fast and reliable, the column stopped making sense.
This was a great read. I think I’m on board with the shift you’re describing—flowcharts make way more sense than rigid boards in a nonlinear world.
That said, I’m curious how teams actually operate inside a system like this. Are there tools today that support modular, dynamic workflows beyond just visualizing them? You can use something like Miro to sketch it out, sure, but execution is the hard part. Diagrams don’t do the work.
Maybe that’s your point though: remove the human from workflow management altogether and let the tools handle the flow themselves.
Approx. 10 years ago, Roman Müller designed what he called „Spaceship Kanban“ to visualize non-linear flows:
https://roman-mueller.com/2017/06/08/spaceship-kanban/
The example shows that it is possible and in a fun way too!
Thanks for this piece, Jurgen.
I'm not sure I understood what "non-linear" means here. The fact the steps can appear and disappear or that the work doesn't always move 'left to right' on a kanban board makes it non-linear?
If that is the case, I would suggest that the work is still done in a linear fashion - it's just that what is worked on and when (and by whom) is dynamic. Being able to respond to this dynamism efficiently is definitely beyond the capabilities of typical kanban.
One thing we would do on my teams is account for the feedback loops by simply creating another work item (e.g., I found a bug, the bug is a new Issue with dependencies and effort and needs to be sequenced in) and accounting for the time and schedule impact using my Agile Projector tooling. The fact that it needed a new visualization was not a reason to stop visualizing it, but rather a reason to make re-visualizing easy. We also built metrics to measure this variance, with the goal of minimizing it (when economical) through better planning and ownership in the flow or just naturally accounting for it in the +/- of schedule projections.
For AI, I prefer to ask the question - is this a deterministic process often enough (e.g., once a day) that we are really looking at a rules engine and good tooling vs. a need for AI? It seems that a list of Issues that is updated once a day with size and dependency for each, and a team with customizable skills, preferences, availability and velocity doesn't need AI to create an optimized schedule.
I agree that it doesn’t make sense to visualize complex work in a linear way. Intrinsically it already must be wrong. I don’t think that it is valuable to visualize the exact process steps for such work. I wouldn’t even try. I think abstraction (fuzzy) is a way to avoid going down the rabbit hole and constantly updating boards or flow diagrams. In the end a board is an information radiator. I guess the information needed depends highly on the context.