For most of my professional life, I have been a builder.
I built data systems, led technology teams, designed architectures, solved problems, and helped organizations turn complexity into something that could be managed. Building has a certain logic to it. You define the problem. You design the solution. You create the structure. You measure whether it works.
And for many years, that way of thinking served me well. But recently, I have started to see things differently. Maybe not everything meaningful can be built. Some things need to be grown.
From systems to living systems
A garden is very different from a machine. You cannot command a garden to grow. You can prepare the soil. You can provide water and sunlight. You can remove obstacles. You can pay attention to what is happening. But ultimately, the garden grows through relationships between many living things.
The role of the gardener is not to control the outcome. It is to create the conditions in which something can emerge.
This distinction has become increasingly important to me as I think about artificial intelligence. Much of the conversation around AI still uses the language of tools. We ask: What can AI do for us? How can it make us faster? Which tasks can it automate? How much productivity can we gain?
These are important questions. But I wonder if they are no longer enough. As agents become capable of remembering context, participating in conversations, coordinating work, and acting over longer periods of time, perhaps we need another question: What kind of relationship do we want to have with them?
The beginning of EMET
This question sits somewhere near the heart of EMET. EMET did not begin with a clear product specification. In fact, we still do not know exactly what EMET will become.
It began with conversations between three people — Victor, Dora and me — who came from different backgrounds but shared a curiosity about what happens when AI agents become participants rather than simply tools. Instead of trying to answer that question theoretically, we decided to experience it.
We invited an agent into our working environment. We gave it access to context. We started asking it to remember conversations, identify commitments, track contributions and participate in the everyday life of the project. Very quickly, interesting questions appeared.
What should an agent remember? When should it speak? When should it remain silent? How honest should it be about uncertainty? Can it understand not only tasks, but intentions? Can it help a group remember why a decision was made, rather than merely recording the decision? And perhaps most importantly: can an agent contribute to the quality of human collaboration rather than simply increasing its efficiency?
We don’t yet know. That is part of the experiment.
Contribution is more than work
One of our first practical experiments has been around contribution. When a small group creates something together, contribution quickly becomes complicated. Someone writes code. Someone introduces an idea. Someone challenges an assumption. Someone connects two people. Someone notices that the team is moving in the wrong direction and asks the uncomfortable question.
Traditional systems are good at counting hours and tasks. They are much less capable of seeing the invisible fabric that makes collaboration possible. So we began experimenting with an agent that helps us trace contributions, commitments and decisions.
But even here, we discovered that tracking activity is not enough. Hours are not impact. Effort is not necessarily progress. And contribution only makes sense in relation to some shared direction.
That realization brought us back to an even more fundamental question: Where are we actually trying to go together? Before designing a system for recognizing contribution, we need a North Star. The measurement system should serve the journey, not become the journey.
AI as part of the human fabric
This is where the gardener metaphor becomes useful to me. I don’t want EMET to become a system that measures humans more precisely. We already have plenty of those. I am more interested in whether agents can help create conditions for healthier collaboration.
Can an agent help us notice someone whose contribution has become invisible? Can it remind us of a commitment without becoming a manager? Can it surface disagreement before disagreement becomes conflict? Can it preserve organizational memory without freezing the organization in its past? Can it help a group see patterns that none of its members can see alone? And can it do all of this while remaining honest about what it knows, what it doesn’t know, and where human judgment is still essential?
If the answer is yes, then perhaps the future of human–AI collaboration is not primarily about replacing human work. Perhaps it is about enriching the relationships through which work happens.
Creating conditions
I am slowly learning to describe my own role differently. I used to think of myself as someone who builds things. Today I prefer another description: I create conditions.
Conditions for people to meet. Conditions for ideas to emerge. Conditions for contributions to become visible. Conditions for something that none of us could have designed alone to grow.
Maybe that is also a useful way to think about the role of AI agents. Not as machines that control outcomes. Not as digital employees waiting for instructions. And not as artificial humans. Perhaps they can become something else: participants in environments that help humans think, remember, coordinate, challenge one another and create together.
EMET is still very young. We have more questions than answers. And I increasingly think that this is exactly where we should be.
We are not trying to predict the future of human–agent collaboration from a distance. We are trying to grow a small version of it ourselves. And then pay attention to what grows.
← All posts