x + 3 -

In praise of middle management

There’s only so far you can scale with one person directing a team of juniors. Good management is our current bottleneck in scaling up AI agents.

Inspired by a recent post from Sam Altman (2025-02-09):

We are now starting to roll out AI agents, which will eventually feel like virtual co-workers.

Let’s imagine the case of a software engineering agent, which is an agent that we expect to be particularly important. Imagine that this agent will eventually be capable of doing most things a software engineer at a top company with a few years of experience could do, for tasks up to a couple of days long. It will not have the biggest new ideas, it will require lots of human supervision and direction, and it will be great at some things but surprisingly bad at others.

Still, imagine it as a real-but-relatively-junior virtual coworker. Now imagine 1,000 of them. Or 1 million of them.

Having had some “real-but-relatively-junior” coworkers, I’m familiar with how much supervision and direction they need to contribute to a product/company/team. I invest X units of my time in them; they produce X * Y units of output. If all goes well, Y > 1; reality often falls short.

There’s only so far you can scale with one person directing a team of juniors. I can work with a few; some other engineers may be able to direct a few dozen. Eventually, you get to the point where you’re spending all of your time supervising and none of it producing that output yourself. Congratulations, you’re a manager. (Yes, this is a gross oversimplification of the manager’s role.)

Is it productive for a CEO to directly supervise and direct a thousand (or a million) junior employees? I imagine few (if any) would think so. So our CEO hires managers, who hire managers, who hire managers… who supervise and direct a reasonable handful of junior employees (often mixed in with senior employees who can take an active role in mentoring the juniors). These managers coordinate their teams with all the other teams to achieve the company’s vision, with the coordination challenges growing geometrically as the company scales.

I’ll not predict whether/when AI may eventually replace senior employees or managers. But until it does, good management is our bottleneck for scaling up our “real-but-relatively-junior” armies of virtual coworkers.

European Tour

Several years ago, Modern Tribe invited me on the annual team trip. Instead of the “usual” Central American/Caribbean destination, we were meeting in Tuscany. After the trip, Stephanie would join me in Rome, where we would have a few days to explore before heading home. This adventure was scheduled for May 2020—it didn’t happen.

Jump forward four years. The kids are teenagers, finances are more flexible, the world is open again. We decided it’s time to try again, this time as a big adventure for the whole family. In May 2024, we went on the Brinley Family Grand European Tour, a 17-day trip with stays in Rome, London, Paris, and Munich.

I will not herein attempt to capture the entire journey. Rather, I want to highlight a handful of experiences and create a space to share selected photographs. (Stephanie has her own selection of photos over at Laughter & Dance.)

Notable Sights

The Colosseum, which apparently had a retractable roof

Rome is full of “nasoni”, fountains dispensing potable water

Musicians have to compete for space with storage carts in Westminster Abbey

Liechtenstein might be the most beautiful place in the world

Great Food

Everything was delicious. Our first evening in Rome included a food tasting tour, and a few friends recommended restaurants. Otherwise we just searched Google Maps for nearby restaurants with 4.5+ ratings.

Magnificent Organs

We encountered a plethora of churches/cathedrals/chapels filled with beautiful organs, although we only had the opportunity to hear two of them: in St. Paul’s Cathedral for the Sunday eucharist, and a short concert in the Salzburg Cathedral that featured three of the seven(!) organs therein.

Odds & Ends

How to reduce latency with Samsung Galaxy Buds on macOS

I use Samsung Galaxy Buds as my primary headphones. They’re great. Sometimes, though, I notice that audio coming from my Mac is delayed by ~1 second compared to the video.

After a bit of digging, I’ve discovered that macOS defaults to using the SBC codec, which does some buffering that can lead to that delay. Switching to the AAC codec (or fiddling with SBC settings) can eliminate that delay.

The magic incantation:

sudo defaults write bluetoothaudiod "Enable AAC codec" -bool true

Reconnect your headset, and the delay is unnoticeable.

Organize SSH config with Includes

My SSH config file has grown organically over the years. Overall, I try to keep it organized and well-commented. Today I thought to ask if I could include other files in my config, to separate groups of related servers into separate files. The answer, delightfully, is yes!

You can include specific files, or even an entire directory. I went with the latter:

Include config.d/*

Adding this line to the top of ~/.ssh/config means that I can create as many files as I want in ~/.ssh/config.d/, and any Host stanzas in those files will be included as if they are in ~/.ssh/config.

Now I have a separate file for each customer who’s servers I access: ~/.ssh/config.d/customer1, ~/.ssh/config.d/customer2, etc. Any time I add a new customer’s servers, I can add a new file and it’s automatically included. This makes it easier both to see which servers are associated with a customer and to see which customers’ servers I may have access to (for when I need to clean up later).

Where the Sidewalk Ends

Software developers work on the border between problems that have already been solved and the uncharted wilderness of possibility.

In my work as a software developer, I solve problems. Specifically, I tend to solve problems that require me to write code that interacts with other systems, frameworks, APIs, etc., extending them to do things that they were never designed to do. Some of those systems are designed to be extensible, others… less so. And it’s typically in those hard-to-implement edge cases that my skills are most called upon.

I work on the border between problems that have already been solved and the uncharted wilderness of possibility. Living in this rustic environment, I spend most of my day staring at flaws and limitations. If only this system exposed the data I need; if only that API didn’t have a jarring inconsistency; if only I could trust the value returned by this method call. But these things exist as they are, and it’s my job to find ways around them. Despite some internal grumbling, it’s usually quite enjoyable work.

Many of the products and platforms I’m trying to extend were designed with particular audiences in mind. Customers in the target audience have common problems that can be solved with common solutions. The problems I’m trying to solve… these are not the common case. When a product successfully anticipates the needs of a customer, when it solves the customer’s problem with minimal effort, when it’s exactly the right tool for the right job… I never see these projects. They’re someone else’s responsibility, because they don’t require my particular skills.

As I poke and prod at these systems, as I uncover the edge cases that push them to their limits, I need to remind myself that my challenges and my frustrations are not indicative of the quality of the product. A product that solves most of a problem for some percentage of its audience (whether that be 95% or 5%) is to be applauded. It’s doing its job well for those users. Even if it’s not for everyone, even if it doesn’t solve every problem, for some users, this is just the solution they need. For everyone else… it’s a big world; there’s room for another product in the market.