When Building Gets Faster, Understanding Matters More

As AI changes what smaller teams can accomplish, what might it mean for how we design organisations, distribute responsibilities and enable the flow of work and value creation?

For years, we've talked about stream-aligned teams as relatively small, cross-functional groups of people, typically around five to nine, bringing together the skills and capabilities needed to deliver value to their users.

There are good reasons for this. Teams need enough people to bring together different perspectives and skills, but not so many that communication and coordination become overwhelming. We've also learned the importance of giving teams clear responsibilities, manageable cognitive load and enough autonomy to make decisions and deliver without constantly relying on others.

But AI is changing what's possible within those boundaries.

A team of three or four people, working effectively with AI, may increasingly be capable of work that previously required seven or eight. AI can extend the range of skills available to a team, make unfamiliar technical knowledge more accessible and reduce the effort involved in implementation, testing and maintaining software.

This raises some interesting possibilities for organisational design.

Could smaller teams take responsibility for more complete capabilities? Could easier access to specialist knowledge reduce some dependencies between teams? Might we reconsider where stewardship boundaries sit, or how platform capabilities are provided and consumed?

And perhaps more fundamentally, if the capabilities of individual teams are changing, should we be reconsidering how we organise the work around them?

These are questions I've been exploring through the lens of User Needs Mapping. Rather than starting with an assumption about the ideal size or composition of a team, we can begin by understanding the users we're serving, the capabilities required to meet their needs and the interactions necessary to deliver value.

That gives us a way to explore the opportunities AI creates, while understanding the consequences of the organisational decisions we make.

What Does Autonomy Look Like When Teams Can Do More?

One of the most interesting possibilities is that AI could allow smaller teams to take on responsibilities that previously needed more people or specialist support.

But autonomy isn't simply a function of how much work a team can complete. It also depends on the boundaries around that work, the knowledge available to the team and the relationships it has with other teams.

A small team with responsibility for a coherent capability, clear interfaces and accessible organisational knowledge may be able to accomplish considerably more than we'd previously have expected.

Conversely, increasing a team's implementation capacity won't necessarily help if its responsibilities remain fragmented across multiple teams.

Understanding those distinctions creates an opportunity to rethink how capabilities are grouped and stewarded, rather than simply making existing teams smaller or expecting them to produce more.

And that's where understanding the wider organisational landscape becomes particularly valuable.

Understanding the Landscape

One of the recurring challenges I encounter is that people often understand their immediate area of responsibility reasonably well but have less visibility of how their work connects to the wider socio-technical system.

Who are we serving? What do those people need? Which capabilities contribute to meeting those needs? Where do responsibilities overlap, and where are teams dependent on one another?

These questions are central to User Needs Mapping.

Consider a fictional travel booking organisation, Acme Travel. Travellers want to find suitable trips, book them and manage their bookings. Customer service agents need to resolve booking problems on their behalf.

Meeting these needs involves several capabilities, including booking, pricing, identity, customer accounts and notifications. Some contribute to multiple user needs, while others depend on shared capabilities stewarded by different teams.

A User Needs Map makes the relationships between users, their needs, organisational capabilities and team stewardship visible. **Created using Fast Flow Toolkit.

The map gives us a shared view of the organisational landscape. We can see which capabilities contribute to particular needs and begin to understand how responsibilities are distributed.

This matters when considering smaller, AI-enabled teams. Their ability to build and change software more quickly doesn't remove the need to understand where those changes belong or how they affect other capabilities.

Without that understanding, we risk accelerating the creation of duplicated functionality, unnecessary dependencies and complexity that someone else will eventually have to manage.

Understanding the Interactions

Of course, a User Needs Map doesn't tell us everything about how work moves through an organisation.

We also need to understand how teams interact to make changes.

In our Acme Travel example, Booking Experience depends on capabilities stewarded by Customer Platform. Changes to booking integrations require ongoing coordination between the teams, even when the underlying platform capabilities don't need to change.

Customer Operations needs to wait for the Customer Platform to complete work before they can continue (they are temporally coupled), requiring account-related changes to be released in step. There's also a handover from Booking Experience to Customer Operations.

These interactions introduce structural constraints on the flow of change.

A Team Interaction Model makes ongoing coordination, temporal coupling and handovers visible, helping us understand what may be constraining the flow of change. **Created using Fast Flow Toolkit

Imagine that Booking Experience starts using AI extensively. It can generate code, update tests and implement changes much faster than before.

But it still has to coordinate integration decisions with Customer Platform and hand work over to Customer Operations. Customer Operations still faces coupled releases.

The implementation work may have accelerated considerably, while the overall flow of change improves very little.

That doesn't mean every dependency or interaction is a problem. Some coordination is necessary and valuable. The important thing is understanding the nature of those interactions and whether they're helping or constraining the organisation's ability to meet user needs.

This is where I think AI creates an interesting opportunity. Rather than simply increasing the amount of work individual teams can complete, we could use this increased capacity alongside a better understanding of the organisational system to improve how work moves through it.

From Understanding to Organisational Decisions

Making the landscape and its interactions visible is useful, but it isn't the end goal.

We want to use that understanding to decide what should change.

Returning to Acme Travel, suppose we investigate the repeated coordination between Booking Experience and Customer Platform.

We discover that many booking changes require both teams to be involved, even when the shared customer account and identity capabilities remain unchanged. We also find that neither team holds the complete picture of how booking flows integrate with those capabilities.

This suggests the difficulty may not be the dependency itself, but an unclear boundary around stewardship of the integration.

One possible response is to give Booking Experience responsibility for its integrations with Customer Account and Identity, while Customer Platform continues to steward the shared capabilities, interfaces and contracts.

The intention is to allow Booking Experience to make routine changes independently, without requiring Customer Platform to participate in every implementation decision.

But this remains a hypothesis. Moving integration responsibilities could increase Booking Experience's cognitive load or allow different consumers to diverge in how they use shared capabilities.

We need to make those assumptions and trade-offs explicit, then evaluate whether the change actually improves outcomes.

Connecting observed problems to a proposed structural change, with explicit assumptions and measurable outcomes. **Illustration created based on an Acme Travel Flow Decision Record from the Fast Flow Toolkit.

In this example, the desired outcome is for booking changes to reach travellers the same day. Two measures help us evaluate progress: reducing booking-change lead time to one working day and reducing the proportion of changes requiring Customer Platform involvement to 20%.

This gives us something more useful than a general ambition to make teams faster. It connects a specific organisational decision to the needs we're trying to meet and provides a basis for learning whether the intervention has worked.

And that remains important regardless of how much AI improves our implementation capabilities.

Organisational Understanding as Context for AI

There's another dimension to this that I find particularly interesting.

AI is increasingly capable of helping people navigate unfamiliar systems, explain technical dependencies and recover knowledge that might previously have been held by a small number of individuals.

That could reduce some of the risks associated with knowledge being concentrated in particular people. But retrieving information about a system isn't the same as understanding why it exists, what its users need or whether its current design makes sense.

Code and technical documentation tell us something about how a system works. They don't necessarily explain why a particular capability exists, which user needs it supports, or why responsibilities are distributed across teams in a particular way.

What if we could make that organisational context available to AI as well?

A structured understanding of users, needs, capabilities, teams, dependencies and decisions could help AI reason within a more meaningful organisational context, rather than working from disconnected technical artefacts.

It wouldn't make the model infallible. Organisational maps are representations of our current understanding, and they contain assumptions, omissions and relationships that need to be validated.

But that context could help us ask better questions, constrain proposed solutions and understand the wider consequences of changes.

The value isn't simply in documenting the organisation. It's in making our understanding explicit, so that people can challenge it, improve it and potentially use it to guide AI-assisted work.

The Economics of Building the Wrong Thing

As producing software becomes easier, it also becomes easier to produce software we don't need.

We can spend time and tokens exploring solutions, generating code and refining implementations before establishing whether we're addressing the right problem.

And the cost doesn't end when the code has been generated. Someone still needs to review it, integrate it, operate it, maintain it and understand its consequences.

The cost of generating a solution may be falling, but the cost of unnecessary complexity doesn't disappear with it.

Understanding user needs and organisational capabilities therefore becomes an economic consideration as well as a design concern. It helps us focus effort on changes that matter, avoid building functionality in the wrong place and make better decisions about where responsibilities should sit.

The faster we can build, the more valuable it becomes to establish what needs building in the first place.

What Changes, and What Doesn't

AI may enable smaller teams to accomplish more. It may make specialist knowledge easier to access, reduce the effort involved in implementation and create opportunities to rethink how responsibilities are distributed.

But smaller doesn't automatically mean more autonomous, and faster implementation doesn't automatically mean faster delivery.

We still need teams with a clear purpose, manageable cognitive load and meaningful connections to the users they serve. We still need to understand dependencies, minimise unnecessary coordination and create fast feedback loops that allow us to learn whether we're delivering something valuable.

AI changes some of the constraints we're designing around. It doesn't remove the need to understand the system.

Perhaps the most important opportunity isn't simply to make existing teams more productive. It's to use this changing landscape to reconsider how capabilities, responsibilities and interactions should be organised so that work can flow and value can be created more effectively.

As the cost of producing solutions falls, the importance of understanding the problem and the surrounding system increases.

Are we using AI simply to accelerate the work we're already doing, or are we using the opportunity to improve how the whole system meets user needs?

I'm particularly interested in hearing from people experimenting with smaller, AI-enabled teams. Are you finding that the boundaries around teams and capabilities need to change, or are established approaches to team design still serving you well?

I'd welcome your thoughts and experiences.

** The Acme Travel example is fictional. The User Needs Map, Team Interaction Model and underlying Flow Decision Record were developed using Fast Flow Toolkit, a tool I'm building to help make organisational structures, dependencies and flow decisions visible.

Next
Next

What Actually Flows Through a Value Stream?