Sol is like typical senior engineer - 0 ability to make trade-off decisions on work allocation, will spend forever chasing edge cases that will never happen, will be proud that everything is "correct", but will ignore that deadline for delivering was 3 months ago.
Krzysztof Cieslak
@kcieslak.io
Artist 🎨 | Photographer 📷 | Globetrotter 🌍 | Speaker 🤹🏻♂️ | Author of Ionide 🚀 | Researcher @ Github Next 🔬| Co-creator of GitHub Copilot, Copilot NES, Agentic Workflows, and Ace | Building the future of software development 💻
I think I'm gonna run experiment to see if i can use iPad as my only daily driver... probably will need to build some software first, but that's what are agents for.
I’ve written blog posts about starting to blog again… bit meta, eh? kcieslak.io/Restarting-t...
Restarting the blog :: Krzysztof Cieslak
Restarting the blog - let’s go on a new adventure
kcieslak.io
Social media and internet are so weird nowadays - half of it is just LLM generate bullshit, other half is wanna-be influencers trying to do self-promotion.
I would say something about Steve but why give him even more free marketing 🙃
there are many people who work in the ai coding field who are keeping up with the almost daily changes, able to give a good review of where things are and where they're going, and where the industry is headed. steve yegge is not one of those people.
AI Twitter seems to be absolutely wild place... not in a good way. Whole space is driven by influencers and marketing shills rather than builders doing cool stuff.
Fable, Sol, Opus 5, Opus 4.8 are all great models for software development. And differences between them are not really huge - your tooling choices, having fast and informative feedback loop, way you structure tasks will have way bigger impact on the results than model choices.
Opus 5 is pretty great model to use right now. Is it as good as Fable and Sol? Nope. But it's good enough while being way faster (especially in Fast mode). For software dev, value of having additional iterations in same unit of time is tremendous as we tend to have really good feedback cycles
150,000 marched yesterday in London for trans rights. Opened the Sky News app. Scrolled and scrolled. Nothing. BBC News app. Nothing. Guardian app. Nothing. This is what actual silencing looks like.
Anyone has good agent skill for handling mdx files? Asking for a friend…
It seems I’ll have busy autumn. Already 3 conference talks planned: * Poland in September * Norway in October * Germany in November And there’s still potentially some more stuff happening. I will share more details as soon as conferences actually announce speakers ;-)
Universe is teaching me a lesson. Never again.
There’s first time for everything - first time in my life I’m rooting for England.
No idea why Tuchel did so defensive substitutions… full gave away control over midfield
There’s first time for everything - first time in my life I’m rooting for England.
It’s first time in many months I’ve switched to OpenAI model as main driver. Is it because Sol is better than Fable? No, I actually don’t really care. They both seems like a good models… I just can’t use Fable due to Anthropic’s data retention requirements.
I miss being young. Fairly sure getting back to work after… uh… “conference” was easier back then 🙃
Next year will be year of formal verification! And functional programming! And Linux on Desktop! - every nerd for last 20 years
One thing that I no longer hear much talk about: Whether AI would help with formal verification go mainstream. Formally verifying programs is pretty hard and is a niche skillset! Doesn't seem like AI has made a difference in that field. Worth asking: why?
Erling Haaland's best friend on the England national team is Jude Bellingham. #WorldCup
Humans will be always most important part of software development.
Agree with the last sentence, and I'm finding Claude in particular much better at producing sensible #fsharp code than it does C#.
Since we can generate as much code as we want one could think that programming languages and ecosystems doesn’t matter (“just rewrite everything to X”)… But I think reverse is true. Good languages, good compilers and good tooling is absolutely huge value for agents…
Since we can generate as much code as we want one could think that programming languages and ecosystems doesn’t matter (“just rewrite everything to X”)… But I think reverse is true. Good languages, good compilers and good tooling is absolutely huge value for agents…