The Art of Delegation: Orchestrating Jules and Claude Day to Day

There is something deeply strange and yet familiar about watching two artificial intelligence agents — Jules and Claude — collaborate on a codebase while my youngest daughter sleeps in the room next door. As a software engineer, automation has always been the holy grail; as a father, delegation has become a survival necessity. But the intersection of these two realities revealed an unexpected complexity: that the true difficulty lies not in making the machines work, but in knowing how to supervise them without suffocating them.

In the ontology of my daily life, every task I delegate to Jules (generally focused on engineering tasks and rigorous scaffolding) or to Claude (more attuned to synthesis and lateral thinking) is not a mere command executed in a vacuum. They are events. And, as a process philosopher would argue, they are events all the way down. Each prompt is a perturbation in the system, an encoded intention that reverberates through latent space until it returns as code, text, or architecture.

The Paradox of Control#

The hardest lesson I learned in the last year orchestrating these agents did not come from a syntax error or an infinite loop, but from my own psychology. The engineer’s initial impulse is micromanagement. We want to review every line of code the moment it is generated, correct every comma, guide the invisible hand of the machine.

However, I discovered that treating autonomous agents as mere extended keyboards destroys the very advantage they offer. It is like trying to teach a child to walk by holding their heels. The magic happens when you define the contours of the problem, the success criteria (the “schema,” if you prefer), and allow the agent to navigate the solution space.

But how do you maintain human oversight? How do you prevent the silent hallucination that corrupts an entire system?

The Practice of Architectural Supervision#

The answer I found lies in a shift of posture: I stopped being the executor to become the architect and the editor.

When I trust Jules to refactor a microservice, I don’t look at the code while it writes it. I build rigorous tests and CI/CD pipelines that act as the laws of physics of that particular universe. Human supervision is transmuted into systemic constraints. If Jules’s code compiles and passes the integration tests I designed, it has the freedom to be idiosyncratic in its implementation.

Claude, on the other hand, acts as my system design partner and philosophical sounding board. When we discuss the intelligibility of a new architecture, my supervision is rhetorical. I apply Socratic skepticism, forcing the model to justify its structural choices against the practical constraints of the business and long-term maintenance.

Parenting and Process#

There is an undeniable parallel between managing these agents and raising children. In both cases, you are dealing with entities that possess a degree of autonomy, that interpret your instructions in unpredictable ways, and whose “output” does not always correspond to your original intention.

When my daughter builds a tower of blocks and it collapses, my role is not to rebuild it perfectly. It is to help her understand gravity. In the same way, when an agent proposes an overly complex design pattern, my role is to redirect its attention to first principles.

Events All the Way Down#

In the end, orchestrating AI is an exercise in epistemic humility. It is accepting that we cannot (and should not) have absolute control over every atomic operation of our systems. We must focus on boundary conditions, interfaces, and fundamental guarantees of stability.

Effective delegation to AI is not the abdication of human responsibility. It is its elevation to a higher order of abstraction. As I watch the terminal run the test suites written by Jules, and Claude’s text take shape on the screen, I realize we are not building tools that replace us. We are building a new kind of collaborative system, where human discernment and the probabilistic brute force of the machine converge in a complex and continuous dance.

The screen glows in the dark study. The baby breathes softly through the baby monitor. And the events keep unfolding.