Delegation for Humans or AI is a Continuous Function

I originally wrote this blog post years ago when I was running Metamor, a consulting company with over 500 employees. My goal was to help my new managers lead their teams more effectively.

As I build Backtick, an AI-first company, I have come to realize that the same concept applies to AI agents. AI agent implementations work best when you treat them the same as employees. As Ronald Reagan once said in another context, "trust and verify".

In electronics, signals are either analog or digital. Analog signals are smooth and continuous, while digital signals are composed of small, discrete steps. Think of the difference between a staircase and a ramp: on a staircase, you are either on one step or the next, but on a ramp, you can exist at an infinite number of points between the bottom and the top.

When it comes to delegation, whether you are managing a human employee or an AI agent, it is a mistake to view the process as a digital, "on-off" switch. Many managers mistakenly believe that once they assign a task, the responsibility has fully transferred to the other party. However, because the outcome remains your responsibility, delegation is actually an analog function. You are not handing off a task; you are sharing the responsibility for it.

To succeed with an AI agent, you must treat it much like a traditional team member. Just as you would with a human, the quality of your results depends entirely on the quality of your instructions. An AI is not a "set it and forget it" tool; it requires the same level of clear communication, context, and expectation-setting that you would provide to a new hire.

Once the task is assigned, the work is not finished. You must manage an iterative process of checking, refining, and guiding. Whether you are working with a person or an AI, I recommend a continuous approach to ensure effectiveness:

After assigning a new task, establish clear checkpoints based on milestones rather than arbitrary time intervals. For an AI, this might mean reviewing the output after the initial data retrieval or the first draft of a summary; for a human, it could mean checking in after completing a project outline or a preliminary analysis. As you gain confidence in the agent’s or employee’s performance, you can gradually increase the distance between these checkpoints. Move from reviewing granular, step-by-step progress to evaluating larger project phases, and eventually, to assessing only the final deliverables once the process is fully refined and reliable.

By treating delegation as a continuous, analog function rather than a binary "I own it" or "they own it" handoff, you create a feedback loop. This approach allows you to catch errors early, adjust the trajectory of the work, and maintain high standards. Whether you are building trust with a human colleague or fine-tuning the parameters of an AI agent, this continuous oversight limits risk, improves quality, and maximizes your long-term effectiveness as a manager.