Forced to Manage

Read Time: 3 minutes (546 words)

I had a discussion with a friend the other day, and an example made there brought some clarity to my thinking about LLMs and how they’re currently being used in dev workflows.

A significant number of developers that I interview describe their dev workflow as, effectively, managing the LLMs as if they were a team of junior engineers. The developer does the planning and architecture work, perhaps with collaboration from an LLM agent, and then sends out the tasks to individual coding agents to do implementation. Finally, the results are aggregated and the developer reviews it all before signing off on it.

Ignoring a large amount of details that I’m glossing over, and a decent amount of individual variation between developers, this is the core workflow. Plan the work, farm it out to LLMs, review the work.

There are two problems with this, and both contribute to why I dislike the idea of even using LLMs in my job:

1. I Enjoy Doing the Work

I got into software development because I like writing code and solving problems. Outsourcing the coding to something else may be more efficient (not as often as the hype would make you think), but that removes a significant amount of the enjoyment I get out of the job. “Press button, wait 2 hours, have maybe solution” does not trigger my dopamine receptors. The effort being put into the work is where I derive a decent chunk of enjoyment.

And if I’m doing my job correctly, I still have to go through the hard mental work of understanding the problem and devising a solution, and then you add on the extra work of evaluating whatever the LLM output and understanding the details of whatever that happens to be. I have to put in all the mental effort anyway, and realistically only avoid having to type out the code.

2. I Explicitly Do Not Want to be a Manager

I’ve done every part of the SDLC, as both an IC and as a manager. I’ve done the team lead role, I’ve done the department manager role. I’ve played oversight on and mentored devs so junior that they didn’t know what print_r was in PHP. I’ve managed local and overseas teams concurrently. So I can say with very high confidence that I truly do not want to manage people.

I can do it. I’ve got the skills, won from years of pain and toil. But I’m not good at it, and I sure as frak don’t enjoy it. I’m far better as an IC, possibly as part of a larger team, focusing on individual problems and collaborating with others.

Based on how the devs with the most rigorous genAI workflows describe things, I absolutely want nothing to do with managing a fleet of LLMs. A bunch of junior-tier engineers that require constant monitoring to ensure they don’t go off the rails and into the weeds, generate code that may or may not be what will actually solve the problem at hand, require me to constantly clean up their messes, and produce an ultimate result that is far more difficult to read, understand, and maintain than something produced by a human expert? All while sacrificing my own hands-on skills due to skill-drain?

Hard pass, thanks.