Agile for the Age of AI
Do less work. Create more value.
AI can research, analyse, write, design, code, test and coordinate at a speed no human team can match.
That does not make our future managing armies of agents.
The point is not to manage more agents. It is to manage less work.
Building has become cheap.
Building the right thing has not.
The bottleneck has moved from execution to direction, judgement, coordination, context and learning.
So we are uncovering better ways of building by valuing:
That is, while output, execution, coordination and speed still have value, they are means.
The outcome is the point.
Six principles for teams that build with AI
Manage destinations, not directions.
Make the outcome painfully clear.
What should be different for the customer, the product or the business when we are done?
Define success.
Define the constraints.
Then let humans and AI figure out the route.
The goal is the job.
Automate the chores. Keep the choices.
Give AI the work you would rather not do.
Do not protect work simply because humans used to do it.
But do protect the decisions that shape what gets built.
AI should absorb the execution. Humans remain responsible for judgement.
Context over prompts.
A clever prompt cannot compensate for missing context.
Who is the customer?
What are we trying to achieve?
What have we already decided?
What constraints are real?
What does the product already do?
What does the code say?
What happened last time?
Context should live with the work, not inside somebody’s memory, twenty open tabs or an old chat session.
Humans and AI should work from the same living understanding.
Generic context creates generic work.
Builders over roles.
Product can prototype.
Engineering can research customers.
Design can analyse data.
Sales can build tools.
AI makes the boundaries between traditional roles more porous.
Expertise still matters. Deeply.
But expertise should accelerate the team, not become a queue everyone waits in.
Small teams should own outcomes end to end.
Your role describes your strength. It should not define your limits.
Out-learn, don’t out-build.
When building becomes cheap, producing more software stops being the advantage.
The advantage is learning faster.
Did anyone care?
Did behaviour change?
Was the assumption right?
Did the product solve the problem?
What surprised us?
What should happen next?
Output that merely looks right is not progress.
It is inventory.
Reality gets the final vote.
More agency, not less.
AI should make people more capable of changing their environment.
Not less.
More people should be able to take an idea from question to prototype.
More people should be able to investigate a problem without waiting for another department.
More people should be able to build, test, learn and act.
The goal is not to remove humans from meaningful work.
It is to remove meaningless work from human lives.
If you can move the outcome forward, move it forward.
Five commitments
We do not babysit agents.
We pursue goals, together.
We do not optimise for maximum activity.
We optimise for meaningful outcomes.
We do not ask humans to remain the default execution engine.
We use human effort where judgement, creativity, empathy and taste matter.
We do not add AI to yesterday’s process and call it transformation.
Adoption without adaptation is theatre.
And we do not confuse speed with progress.
Shipping faster matters only when it helps us learn faster.
How the work runs
Set the goal.
Make the context available.
Let AI absorb the work.
Keep humans responsible for the choices.
Stay close to reality.
Learn.
Change direction.
Repeat.
Humans decide what matters. AI does more of the work. Together, we learn our way to the outcome.