11 jottings so far.
Reactions to articles, quotes, half-formed thoughts, and things worth writing down but not edited into concise ideas yet.
Maybe my reality would be more aligned with my values if I tried to keep alternative metrics in mind, rather than falling back to the easiest existing metrics like money/net worth:
They are harder to measure, but I think measuring isn’t the point. Just asking myself the question “Is this optimizing for money, or for materially improving human lives?” usually makes things quite obvious.
“Metrics are useful because they compress information…They are dangerous, because they compress information.”
-> source - C. Thi Nguyen, The Score: How to Stop Playing Somebody Else’s Game.
Wealth in the modern era seems to be the privilege of disconnecting without consequence.
is it cool AI lets you build completely bespoke tools quickly? YES! did the majority of the population ever want that? Eh, not really.
Regular people aren’t tinkerers, they don’t want to build their own personal apps. They already had the ability to do some form of this, if they wanted to, with Google Sheets, No Code apps, etc. For the same reason, self-hosters stay a tiny minority. Users will put up with a lot of friction & bullshit as long as they don’t have to tinker. All you need to see is the scary stats around how many people never change the default settings.
What’s my takeaway? The people that are suddenly using AI to build their first iPhone app or startup without knowing how to code are not a sign that AI will give everyone personalized software. I think it’s just people who were already tinkerers in their own domain graduating into a new space to tinker, but not a new swathe of the population getting onboard.
TL;DR: what if PR review was as engaging & low-friction as TikTok?
We’ve had a lot of innovation in AI agents producing more code, but I haven’t run across much in the way of innovation around how we do code review/PR review. The pace of new code is increasing, but humans reviewing are often still the bottleneck.
The knee-jerk reaction I have to get ahead of is “jUsT hAVe AI aGEnTs REviEw IT” - yes, that can be a tool which helps us, but it’s not a solution that covers us 100%. For one, many organizations still require a human-in-the-loop and 1-2 human reviewers for each PR, and we have to work with that constraint until Dark Factories become the norm (big maybe there).
I’m starting to explore some of these ideas in my research repo, feel free to take a look at the interactive demos, or read the context page for the distilled summary of my conversation with Claude about it.
Can we make agents more reliable with better deterministic tooling?
I look forward to when tooling around AI agents has standardized to solve the common frustrations. One of those is trust that the reality of what it creates matches the plan we refined together.
The features the agent & I are building get loaded into my mental model of the codebase, and each time I hit a situation where it only applied a concept to 1/2 the places it should have, I lose trust and feel like I have to spend even more time at each stage verifying every little piece. It’s the difference between handing a task to your competent coworker, vs giving the task to that one architect you wouldn’t trust to install a CLI tool by themselves. One of them gets a LGTM, the other I will dig through with a fine-toothed comb.
While there are a bazillion approaches people might say would have solved this for me (cue the BMAD, Speckit, etc crowd), I believe software engineers will eventually settle on a standard toolkit, in the same way most people use NPM. There are other options to provide competition, but we aren’t all building our own package managers.
There’s this trend of Rust rewrites of tools that either make common tasks super fast or simplifed, and then they get significant adoption. It feels like some form of Jevons Paradox when making things run faster & people want to use it more. A few examples:
ripgrep (2016)astgrep (2023)uv from Astral (2024)Probably numerous other instances of this trend continuing, especially now that agents are making it cheaper to pick performant languages like Rust, even if it would have been more complex for the developers in the past.
I am the bottleneck with AI development.
I’ve been taking days polishing & perfecting my design doc for a new project because I want it to start from the “perfect” context. GAH. I’m hanging on to the optionality, since if I haven’t started yet, then if I just add enough AI Skills, enough notes, enough CLAUDE.md instructions, the repo will come out absolutely perfect and not need iteration.
In a similar vein, I’ve noticed that even though LLMs/agents are so powerful and can do so much work for us, I still have the same struggle taking ideas from thought into action. In other words, I haven’t found the current iteration of agent systems to solve the executive dysfunction problem. It’s great that some people can have 10 agents working on all of their ideas in the background at all times, but I struggle to allow even 1 a week.
There’s a Midwest Archetype of couple I see in the Chicago/Milwaukee/Madison area, though it likely could be drawn to cover a much wider swathe:
With modern tools, I could live my whole life in my apartment and not have to interact with any humans. Leases can be signed electronically. I can move in and have a car delivered to me. All my furniture, groceries, and other necessities are delivered silently. I can work remote jobs and keep my camera off at all times.
We’ve made things so convenient that the world doesn’t require you to interact or be part of your community anymore. Ironically by making common tasks more convenient, it seems like we’ve added friction to the thing that is arguably more important for our health: socializing with other humans, which we used to get at least some by default.