My point, and maybe it's overly simplistic, is that at some point you are using the computer to make computer fonts. At some point the ideas go into the computer and you are using computer tools to build the font. They start in the shower or from the finger feel of carving stones whatever, but before and after AI computer fonts are built with computer software tools. Carving rocks, sure, human hands.
The idea that AI is different, in kind, than any of the many other tools that you use is silly to me -- at no point is it desirable for you to throw out your brain. As you say, every tool has a different sort of "grain", which is neither good nor bad but does in fact influence the final product.
Where you lose me is why would you implicitly throw out your process? What about this tool gives the idea that you should not keep the parts of the process that is valuable, on paper etc? AI probably would do better fiddling with the beziers or whatever, what would be the reason that it's ipso facto a bad idea? Who is saying that?
I'm not sure basically what the other side of your argument was supposed to be.
What sort of human hands make computer fonts? Are you there coding in the ones and zeros on some sort of breadboard, wiring up the data using an oscilloscope and rare earth magnets?
There seems to be this idea that using AI is the same thing as walking away and not touching it. Like that you would just say to the computer, do the thing and then not look at it, read it or edit it in any way. What's the world where you're describing where you wouldn't select and edit and refine? You know, with software as an example, how that works with AI is that you spend all day opting it and refining it and you do it over and over and over and you're guiding it and you're shaping it. So it's not like the computer does it. You're still there. I mean, you're making a typeface. You're not chiseling something into stone. You're making a program of sorts that will make shapes. And I don't know why there would be any world where you would just walk away from your first prompt and call it done.
My wife suddenly died 4 years ago, we had 3 little kids at the time. You need to focus on a) being kind to yourself, whatever that means to you and b) the good things about the change. All change has good parts and bad parts, and it's easy to focus on negative things. Especially with a death since in magnitude they are probably greater than the good things. But if you focus on what you lost you will simply lose the rest of yourself.
I also think that women have a harder time with this than men, possibly because maternal death in childbirth used to be so much more common. But this is just a guess. Certainly until it happens to you its not the sorts of things that you think about too much, and once it does happen you tend to speak to people who are going through the more acute phase of it since they are still actively processing it.
I tested 15 different agents and come up with my own way to range them. Would you hire an agent? Does it spark joy? What is the output quality? It turned into a 61-page deep dive into all the nitty-gritty.
From IDE beasts like Cursor and Copilot to CLI warriors like Aider and full-stack champs like Replit and v0, etc – it’s a no-BS breakdown of what these tools can actually do when you throw a real-world web app prompt at ‘em.
All the resulting code is on
So, Who’s Crushing It?
Cursor Background Agent, v0, Warp: These three scored a near-perfect 24/25. Production-ready, polished, and just chef’s kiss. Cursor Agent was like, “Huh, didn’t expect that level of awesome.”
Copilot Agent & Jules: Tight GitHub integration makes ‘em PM-friendly, though they’re still a bit rough around the edges.
Replit: Stupid-easy for casuals. You’re trapped in their ecosystem, but damn, it’s a nice trap.
v0: UI prototyping on steroids. NextJS and Vercel vibes, but don’t expect it to play nice with your existing codebase.
RooCode & Goose: For you tinkerers who wanna swap models like Pokémon cards and run ‘em locally.
Who Flopped?
Windsurf. I wanted to hate it (gut feeling, don’t ask), and it delivered – basic tests, flimsy docs, and a Dockerfile that choked. 13/25, yawn.
Pro Tips:
Software Pros: Cursor + Warp is your power combo. IDE + CLI = dopamine hits for days.
Casual Coders: Replit’s your jam. Zero friction, instant hosting.
Designers: v0 for quick, slick MVPs. Just embrace the NextJS cult.
Tinkerers: RooCode or Goose. Total control, local LLMs, open-source swagger.
The full report’s got the juicy details – screenshots, rants, and all. I will be doing another report on agents at the end of the summer – let me know what’s your go-to coding agent in 2025. Drop your hot takes or grill me on specifics below.
Ruby is really good at writing DSLs, Javascript sucks at it.
I'll happily admin that this is user error, but I always have trouble editing Gruntfiles when I start to bring in other packages. Adding an additional task to a Gruntfile is way harder in than Rake, or make, or even Ant for that matter, mainly because you are dealing with hand editing a huge JSON blob.
Tweaking the file to add a different path in or whatever is fine, but there are many times where I wanted to to copy and paste something in to make it work, and I spent as much time fiddling with the format of the js file as I did solving my problem.
A lot of people love grunt, but I'm not one of them.
There is code in there to pull the arp table and use that, but it's commented out because I never found a good way to determining how long things stay in the ARP cache. Since we don't know how long they stay in there, you will lose all of the "xxx left the network" notices, so we made that trade off.
The reason that I used redis here is because I wanted it to work even if you scheduled the thing in cron, or have it work when you restarted the process and not give any false messages. Also, putting things in redis makes it easier to integrate it with something else, e.g. a UI, without really having to tool anything.
But yes, at this point it doesn't really need to use redis and if you don't already have a redis-server running it's a bit of unnecessary work to get the proof of concept working.
I think what you mean by the first sentence is "most of 2048's fan base enjoys it because it's fun".
Your back and forth strategy doesn't work as you get to the higher point levels in 2048 btw. It certainly may get you to 2048, but not really that much further past that. The fact that the ideal strategy changes as you get to 4096 and 8192 is something that makes the game more fun to me.
If you like Three's better that's great, but like I said, I started with Threes, got addicted to it for a while, and then moved on. I got addicted to 2048 for a lot longer than I did Threes. It's not a matter of one being free or more accessible. 2048 is simply more fun, which is another way of saying it's a better game.
That's funny, I started playing Threes, was addicted to it for a while, and when I found 2048 I never looked back. I find 2048 to be a clearly superior game both in terms of visual clarity and repeat playing. The mechanics of Threes are super annoying, especially the end game swipe when there are no more moves, swipe through the pointless point tallying, click next, etc. etc. etc. The cutesy noises and animations seem sort of unfinished to me.
The max card I've gotten with threes is 384, while the max card I've gotten on 2048 is 8,192. With Threes I don't really care about not getting any better, while with 2048 I feel close to getting 16,384 and still think it's interesting to do so. Threes just doesn't have the staying power, the mechanics are simply more frustrating.
I've seen your argument before -- the big post the Threes guys did had it also, that they felt their game was clearly better. I can understand why they felt that way, but I don't get it. And given the taking over the world popularity of 2048, I'm not sure that the evidence is there to support that position.
Not to nit pick, but uptown starts at 60th and above, and most people would consider midtown to be below that to 34th, then the no mans land, and the village to be downtown i.e. below 14th street. Maybe if you were being a jerk about it you'd say that the numbered streets were "uptown", which would make the line to be houston. What you are saying here is that the east village, lower east side, and soho aren't "downtown" which is crazy.
Couple of points (I'm a cofounder of happyfuncorp)
The box is for happy thoughts. Most people find sex a pretty pleasant topic, even if occasionally juvenile.
We filter out hateful or racist things. (I.e. those aren't happy things.) We also filter out the random letters, though I like to leave in the JavaScript injection attempts.
Jon writes on behalf of himself for techcrunch. As I'm sure you realize, we aren't interested in "locking things down".
We are always looking for good people, as is everyone else. The level of skills and capabilities out there is much lower than the demand, and given how much money is involved it is pretty surprising that we aren't, as a society, as the globalized interconnected species we are, training more people to develop the skills. Especially when the tools are so in reach and you don't need formalized training or get licensed or whatever.
We hire between 2-4 people a month, whenever we find a good fit. We are especially looking for team leads and technical project managers, but if you are good then we find a way to make it work. Engineers who can communicate and lead are the limiting resource.
And if you don't think that "jiggly breasts" qualifies as a "happy thought" on a corporate homepage then it's probably not a good match...