It’s not unlike working with more junior engineers, but when I review a junior engineer’s code and provide constructive feedback, they’re learning and I enjoy that cycle. Getting a model to “learn” without actually retraining it is… frustrating.
I definitely still enjoy the process of designing a solution to a problem with an agent and a team though. So the enjoyment of the job is still there. There are just some intense downsides (figuring out how to review and validate mountains of agent-written code, for example).
Yeah. And I gave the manager thing a go a while ago and did not like it at all.
On the other side, I've largely become a manager of agents, writing instructions and reviewing "other people's" code. I've become more productive, sure, but I don't know if I want to be a 10x engineer by managing 10 agents writing the actual code :(
The flip-side of that is now I'm wrestling with the question of "Do I actually enjoy writing code this way?" On the one side, I've made more progress on long-standing side projects in the last 3 months than I made in the last 5 years.
I was extremely skeptical of the FOMO peddlers, especially early on in the "everyone has to use AI" cycle, but I do kinda get it. If you actually build with it, you realize that it IS another tool, and it IS a fairly significant shift in how software is developed.
The end of last year was a bit of an inflection point. The models started because much more capable of producing non-slop code if prompted and guided well. Can they still spew out slop? Abso-fuckin-lutely. But to dismiss it all as slop would definitely be a mistake
It’s definitely a hard case, but I’ve seen it adapt to a messy codebase with a lot of organic history. Windows is a pretty extreme case there, especially since a huge amount of that code wasn’t in the training data (at least it shouldn’t be ;)). But I certainly wouldn’t just write it off.
But if it’s truly novel, or has vague requirements and poor validation criteria, it’ll merrily go ahead and write complete bullshit.
For me, the thing is: AI is extremely good at adapting well understood algorithms into new solutions. If it’s been done in public before (I.e. it’s in the training data), or you can put enough examples of how it should write code into the context, it will do a pretty dang good job implementing it.
I think you would be surprised how effective it will be at C/C++, especially if you can point it at patterns in an existing codebase.
Aspirate?
Yeah, Hellebuyck was the MVP for sure.
Yeah, I would have preferred 4 on 4, but everyone knew the rules going in. I saw the coach for Team Canada talking about it and his opinion was basically: We all knew the rules and 3 on 3 was just as fair as anything
It was particularly heartbreaking to lose it that way, I will say. But that’s hockey for you. It’s a fast game that can shift in a snap. You can have the momentum and then make one bad play and boom.
The Shawinigan Handshake 😅
It’s 3 on 3!?! I don’t think my anxiety can take that 😬
This binational family is having A TIME 😂
ITS NOT GETTING BETTER 😬
I’m glad it at least yielded a goal before the end of the period 😅. Gonna be an interesting 3rd!
You do Rust things now?
Being an adult is overrated
My understanding is that it’s designed to encourage the penalty killers to be more offensive and push for goal opportunities instead of just always dumping the puck down the ice
Yeah, shorthanded goals end the penalty. It’s unique to the PWHL (as compared to the NHL at least).
Didn’t have enough Royale Tiger Towel Paper Towel Power Play energy
We’ve watched two other PWHL games this season and they felt easier to follow, but they were both Vancouver at home and on CBC and Amazon. We’re watching on YouTube this time and I suspect it’s something to do with the streaming quality on YouTube
Something about this feed feels off to me, it feels harder than usual to track.
Yep. Goldeneyes are the only team in the league who are the primary tenant in their arena and have their logo at centre ice
At least with the Blue Jays in it, it’ll be easier to find here 😂
