@aegir yes! Fucking hell. Winking at the camera making me all weak at the knees.

@aegir omg the guy with the moustache though

I'm starting to believe the only reason same sex marriage was legalised is so we spend more money

@repeattofade right? That and just general municipal maintenance

So I have tracking disabled at a VPN level on my Android phone and as a result I can't use assistant even for basic shit like setting a timer so that's cool

So getting the house insulated hasn't helped as much as I would have hoped with keeping the house cool in summer but I think that's largely because the back of the house gets baked, the windows still let heat in (even with blinds down) and then the hot air gets trapped.

Keeping the loft hatch open has helped a little bit at least

God, this bubble burst is going to be so brutal

@gary_bbgames i don't think so! Thanks for the tip I'll take a look into it

I would still expect an Android dev to review the output in this example though. That's a key part for me. I don't want to get AI to do anything that can't be reviewed. But it does sometimes mean it frees up more time.

So I think o3 is probably best for understanding something, for example using it as an iOS developer on a team to understand how a feature was implemented on Android and could be applied to iOS.

But I would stick with Claude for actually coming up with a plan for refractors and actually making code changes since it seems to be the best middle ground.

I actually tried using Claude to create the plan and o3 to execute the plan but o3 just decided to ignore the plan and create the shortcuts anyway. It's trying to be too smart, in the most annoying way.

So this is what I mean around leaning into it and shaping how people use it at work. I'd rather we were using the right thing first time than wasting energy and resources on something that can never work.

You'd maybe assume that o3 would just be the default but that isn't the case. Claude does seem to be the best for code in what I've seen so far.

In an effort to understand the limitations and strengths of different models at work I've been trying different models to make a plan for the same task with the same prompt and constraints etc.

It makes me laugh just show utterly terrible Gemini 2.5 pro is at this, like don't even waste the energy (literally).

o3 is smartest but in the least useful way. It finds shortcuts that I didn't really want, to make the plan easier to execute. So you have to be REALLY specific about what you want. Probably too specific. Plus it's expensive and slow.

Claude 4.5 seems to be the most rounded. It can spit out a really detailed plan without taking an age to think about it and it seems to do it the most "correct" way without taking any shortcuts.

Wow I feel like absolute garbage today

The impostor syndrome never really goes does it

@thatmanmatt wow this is such a non-article. They're just regurgitating stuff and not actually doing any journalism there.

Also, as a senior, if I try not to fall behind and lean into it I can use my seniority to help shape how it's used. Find what it is (and isn't) good at, and try to find ways to use it in the most productive but least impactful way.

Things like helping people write successful first-attempt prompts. So that people aren't trying to run prompts over and over. It's better to waste energy successfully once than waste it multiple times on something that can never work.

Okay so here's my unpopular opinion about AI tools from a developer perspective. Regardless of my personal feelings about it, it's being pushed at work in a big way.

Everyone around me is leaning into it, for better or worse. It's happening. If I don't also try and learn the best ways to use it, I will fall behind.

If I fall behind and end up losing my job for whatever reason, it will be harder for me to find work.

I have a mortgage to pay and we rely quite heavily on my income. It impacts more than just myself if I can't find work.

So, I've leaned into it. And, for work specifically, I have found it useful in more situations than I was expecting. I would rather this wasn't happening but I don't see it going away any time soon.

@moogal yup! I know that when running on a device it looks impressive at first but once that novelty wears off

My hot take on the glass UI is that it's going to look dated very very quickly. Looking at screenshots just a couple of weeks later and it looks tacky