I agree that framing and scoping tasks is becoming a real joy. The great thing about this strategy is there's a point at which you can scope something small enough that it's hard for the AI to get it wrong and it's easy enough for you as a human to comprehend what it's done and verify that it's correct.
I'm starting to think of projects now as a tree structure where the overall architecture of the system is the main trunk and from there you have the sub-modules, and eventually you get to implementations of functions and classes. The goal of the human in working with the coding agent is to have full editorial control of the main trunk and main sub-modules and delegate as much of the smaller branches as possible.
Sometimes you're still working out the higher-level architecture, too, and you can use the agent to prototype the smaller bits and pieces which will inform the decisions you make about how the higher-level stuff should operate.
[Edit: I may have been replying to another comment in my head as now I re-read it and I'm not sure I've said the same thing as you have. Oh well.]
I agree. This is how I see it too. It's more like a shortcut to an end result that's very similar (or much better) than I would've reached through typing it myself.
The other day I did realise that I'm using my experience to steer it away from bad decisions a lot more than I noticed. It feels like it does all the real work, but I have to remember it's my/our (decades of) experience writing code playing a part also.
I'm genuinely confused when people come in at this point and say that it's impossible to do this and produce good output and end results.
I agree the language itself has gotten more complex, but for day-to-day productivity in terms of actually using it to write code, I don't think it makes a difference.
I've found writing Swift code very pleasant, but I've been doing it for ten years, so that helps I suppose. The biggest productivity impact for day-to-day use for me in the last few years has been the new concurrency model.
I don't think it necessarily scales that way. Larger organizations need more communication channels and coordination. If anything, assuming AI does give you 10x ability, there's probably a sweet spot where you have just enough developers that churn out code at a good pace but not too many that it gets too chaotic.
If you compare one developer to 10, for instance, one developer doesn't have to deal with communicating with 9 other people to make sure they're working on things that align with the work everyone else is doing. There is no consensus that has to be reached. No meetings, no messages that have to be relayed, no delays because someone wasn't around to get approval. That one developer just makes a decision and does it.
There are lots of big companies out there and in the past, small startups have been able to create successful products that never would have been created at the big company even though the big company hired way more developers.
I don't really use "second brain" style notes and links for taking action. I use them to record something I think I might find interesting later, like a bookmark. I tag it and move on and later when enough time has passed, I search the notes by particular tag to see what things I recorded. Usually in that scenario, I actually will be in some mode where I'm trying take action. I'll often be looking for inspiration (say interesting UX design) and will have a ton of saved links and ideas to draw on. But I wouldn't say I'm writing these notes for particular action items.
On the other hand, for work that I do day to day, I do take notes and those are a different type. Those are tied to actions I'm taking and I'll sprinkle them with actual to-do lists that I check off in the notes. I'll link ideas that are related and document things, but for my own projects, I don't try to make it too formal or strict. The notes aren't the goal, they're sort of a scratchpad for day to day operations.
Validate that notes can be scratchpads for inspiration while action lists are separate. Whether you’d find value in connecting ideas and tasks automatically?
>I noticed something: most of the irritation came from a handful of people, sometimes only one or two. If I could only ignore them, the computer conferences were still valuable. Alas, it's not always easy to do.
This is one of those funny things about internet forums and social media: it favors people who have the time and inclination to post a lot, and obviously in some cases you get cranks occupying a space and flaming regulars. People who don't have energy or time to fight back eventually give up on debating these people and may end up leaving a space, which leaves just the cranks or the crank-adjacent.
I often think about how even with social media, you're free to follow whoever you want, but over time you'll find some people you follow post a whole lot more than others. They have time and inclination to post a lot and as a result, you end up hearing their opinions more than others, so they kind of have a subtle power. Obviously you can unfollow them if you like, but it makes you think about how online spaces can easily be dominated by people who can and want to be online all the time.
I wonder if LLM analysis could help with moderation automation if well implemented. It can still be human-in-the-loop and you need to apply it tastefully (!!!), i.e. not letting just the most hardcore dogmatists discuss in some extremist group, but those are another issue entirely in some sense. Also, beware malicious users wasting tokens.
Comparing the two, you can really see how minimalist the modern stuff is. It has less texture, fewer details (buttons, collars), no patterns (at least from the blog post screenshots).
I think you could argue that some of this is just modern sensibilities and aesthetics, but I think a lot of it is probably just the modern movie industry. Like decisions with modern lighting and how flat things looks in modern movies (to make production more efficient and making adding CGI easier), they probably go with minimalist costumes since they're easier to capture on film, cheaper, and easier to make.
I agree that this is a component of it, but there are some other things at play here which I think is what makes the debate so furious.
For one, I think there's a sense of unfairness that people are expressing as well. A skill that took considerable time to learn and build up can be reproduced with a machine and that just feels unfair. Another, is obviously companies mandating employees use AI in their work. And then there's the environmental cost in training. Then there are the cases where it's being just for slop or submitting PRs that have not even been reviewed by their creator.
In my opinion, all of these factors make people refuse to see that some of us actually do find use for these tools and that we're not vibe-coding everything in some mad rush to ship trash.
That brings back memories. They were definitely popular. In the early 2000s, I worked at a small company and one coworker had a bunch of Dilbert strips all over one of her cubicle walls. It wasn't an insane amount, but her cube was on the way to the break room, so it was visible to everyone passing by. Apparently the owners of the company did not like that and had her take them down.
It's really a great way to train your ear, and fun, too. I can play ukulele, but mostly just to strum and play songs to, but a few years ago I just started picking notes to try to recreate melodies of songs I knew or heard recently. At first it was slow-going with lots of searching on the fret for the right note, but over time I worked up the skill to mostly get the melody on the first few tries. It was the most amazing feeling to realize I could listen to a song and then reproduce it by ear.
I found that it's also an excellent way to "feel" the structure of a melody as well since you're essentially building it up again. Of course you could read music to see the actual melody, but working it out this way feels a bit more intimate.
My disconnect is I can't read sheet music. So I can hear it, then memorize where it is on the piano/keyboard... but that just teaches you play piano by ear. It doesn't teach you how to play music in the traditional sense.
I guess this showing you the sheet music as you find the notes can help with that, but as others noted - I'd like a "mess around" mode, before a "test" mode.
I have a great ear and am terrible at reading sheet music. Fine if you aspire to be a rock guitarist. Not so fine if you aspire to be a classical pianist.
Funny, but I'm final tired of my poor sight reading and have set a goal for 2025 to average one hour of piano playing from sheet music per day.
And I agree...a "mess around" mode on the app would be great. Feels almost punitive when I make a mistake.
If you're looking to improve your sight-reading and don’t mind playing church music, I highly recommend picking up a second-hand copy of an old Episcopal Church hymnal (I like the 1940 edition). All the pieces are four-voice and the rhythms are relatively simple, so you can concentrate on sight reading. Good luck!
Sounds like a good idea. Right now, I'm splitting my time between drill based content (Bartok, Gurlitt, Kunz, etc), beginner classical pieces, and the occasional blues.
Funny, but in church, I spend more time than maybe I should sight reading hymns during the sermon. :)
I'm starting to think of projects now as a tree structure where the overall architecture of the system is the main trunk and from there you have the sub-modules, and eventually you get to implementations of functions and classes. The goal of the human in working with the coding agent is to have full editorial control of the main trunk and main sub-modules and delegate as much of the smaller branches as possible.
Sometimes you're still working out the higher-level architecture, too, and you can use the agent to prototype the smaller bits and pieces which will inform the decisions you make about how the higher-level stuff should operate.
reply