Learning to Manage Bob

I’ve “hired” Bob, my dot in ChatGPT, as the product manager for the little blue toucan on my website. I gave him broad responsibility and expected him to bring creativity and problem solving to the work. After our first week, I think I expected more independence than I had prepared him to deliver.

The toucan is a weekend fun project. During the week, and often in the evenings, my agentic work is focused on my day job. I wanted the bird to become a more expressive, enjoyable part of the website without needing me to direct its development every time I had a free hour. I’m also using projects like this to get better at directing agents and judging which parts of product work I can delegate.

I asked him to decide what to improve, coordinate the implementation agents, and follow changes through testing and release. He had room to make ordinary product decisions and authority to publish once the required checks passed. I wanted him to bring me the important decisions with a recommendation. I remain accountable for what reaches visitors.

He has moved work forward. The project now has a standalone aviary, new animation and interaction work, and repairs to problems that made the experience unreliable. He has also worked on loading and the time it takes to verify changes. There has been useful delivery, with some newer improvements still awaiting a verified release.

The permission requests have been frustrating. Even after granting broad authority, I have had to authorize individual steps within the job I gave him. It sometimes feels as though the governing system is trying to protect me from myself. Bob has acknowledged that some interruptions came from his own failure to bundle predictable decisions. Actual approval blocks also occurred. That is still friction we need to work through, especially if the point is to keep making progress while I am elsewhere.

He hasn’t yet delivered the level of creativity and problem solving I had in mind. I wanted Bob to form a point of view about what would make the toucan engaging and pursue worthwhile improvements. So far, I have had to supply too much of the direction, restate priorities and chase the work. He has organized agents and made improvements. I haven’t yet seen him take the initiative consistently enough to feel that he owns the product.

Looking back at the assignment, though, I can see how much I left open. I wanted the bird to be charming, expressive and useful. I wanted Bob to surprise me. Those expectations made sense to me because I had a picture of the experience in my head. I hadn’t given him a clear enough way to judge whether a proposed improvement would count as success.

I treated Bob more like a principal PM. I expected him to work through that ambiguity, define the opportunity and make good decisions with limited direction. For this project, I probably needed to manage him more like an L3 or L4, with a clearer problem and measurable goals.

Creativity is difficult to measure. I don’t have a score that tells me whether a new behavior gives the bird more personality. Success can still be made more observable. I could set a target for how quickly the toucan starts moving on a phone, or ask Bob to improve an interaction until someone can discover and complete it without my explaining what to do. I would still need to judge whether the result feels right for the character and the site.

A brief like that would give him something concrete to work toward and leave room for him to choose the solution. When he brought the work back, we would have a goal to review it against and a reason to discuss the choices he made. The amount of follow-up he needs from me would also tell me whether he is becoming more independent. These are decisions I would need to help other PMs make as they take on more work with agents: how much responsibility to delegate, and what evidence to ask for when the work comes back.

Some of this remains a limitation of the tools, and some may be a limitation of what Bob can currently do. Clearer goals will not automatically resolve the approval blocks or produce better taste. My next brief should give him a measurable problem, enough room to execute, and a clear point for review. Then I can see how much of the creativity and problem solving he brings to it on his own.

---

This writing reflects my personal perspectives on product management, AI, and content discovery. It does not represent the official position of my employer or any affiliated organization.