I’ve seen multiple articles lately making claims that with new AI tools, managers should be able to manage larger and larger numbers of direct reports. The most notable is Meta targeting 50:1 ratios in their new Superintelligence team. This isn’t going to be a dunk on Meta post, because whether or not they might deserve to be dunked on, I have never worked there, and have no interesting insights into what is happening inside that company. I do, however, manage a team of about 30 software engineers at a company that’s all-in on AI adoption, so I’d like to give my view on how the job is changing.

In the past six months, as coding tools have crossed the threshold into generally useful, the job has gotten much harder, and I would argue that in a world of Agentic software engineering, we need more robust management teams, not less. I know — you are shocked and surprised that someone on the internet thinks their job is important!

First — I will acknowledge that there are some great ways I am able to use these tools to be better at my job. By nature, I am easily distracted by the big vision documents, the most critical fires, and other generally shiny objects. I’ve learned to manage this at work, but being able to quickly build tools and workflows to support this has been really great for me. I can much more easily track things like ongoing projects, upcoming deadlines, 1:1 cadences, etc. I definitely think that fewer things slip through the cracks than before I started adopting these tools. It has also helped me be more consistent with communication. I still stubbornly write most things myself, but there are absolutely places where an LLM can help with broad research, document summaries, and fast feedback on a first draft.

The hardest and most important parts of the job have not gotten easier, though. My LLM doesn’t have a deep understanding of each of my employees, and it doesn’t care about their career growth or understand their strengths. It doesn’t track their day to day experience and identify where they are struggling. Those things still take time and effort, and if truly done correctly, occupy a lot of a manager’s mental space. The rough math that Andy Grove used in High Output Management 43 years ago still holds — if you spend roughly half a day each week per report, you can manage 6-8 successfully. At some point beyond that you are just diluting the quality of feedback and coaching you are able to provide each engineer.

The most critical piece, though, is that we’re going through the most dramatic upheaval I’ve seen in my 25 years in software engineering. There is a never-ending stream of new tools that you “absolutely must use or you will be left behind”. And some of them are genuinely really powerful, but it takes effort to sort through them. And if you are a software engineer, you have to listen to a daily feed of fear-mongering news articles and pronouncements that your profession is dead. I don’t want to turn this into a treatise on Agentic Software Engineering, but in a nutshell we’ve shifted a lot of the bottlenecks from code generation to other parts of the software development pipeline. There’s still a lot of hard thinking required, and we have to rework many processes, tools, and mental models that have worked for decades. This is a lot of change, and change is hard. This means every engineer needs more support, not less.

So we are balancing some genuinely useful productivity tools against the need to manage folks through massive change and some fundamental limits of human beings. I would be hard pressed to agree that these even balance out, and my instinct is that the job has actually gotten harder. I believe that this trend towards reducing the number of managers in teams will eventually lead to demoralized and less capable teams. If you manage engineers, has your experience been different, or are you seeing the same trends?