@mitchellh Curious agent had target to improve and it did not to a level expected, 2 things I am wondering what would happen if agent was given goal to match expectations and also what would agent come up with if it was given blank canvas with problem statement with stated goal or open goal
@bcherny I think we should focus on 100% business scenarios coverage rather than code, integration test is more important now, as it’s easier to manage and discover and maintain them. Feedback loop has move to one layer up, as agent will complete the feature
@bcherny I agree, I can see quality going up, less incidents and less back and forth, overtime you get intuition when Claude is likely to make mistake and those are due to incomplete prompt or not able to provide full context, I have done endless reviews of tests written, they are spot on
@DeniseM22572820@lisamarie12o7@ABC IMHO when buddies pays, it’s a favour which will be repaid through other means which would be indirect tax to tax payers, there is no such thing as free lunch.
@greg_0ld@Number10cat@London_W4 Agreed mate , just moaning to express disappointment, we have duty to report it, this is way we can have a functioning society.
@Number10cat@London_W4 What’s the point, my wife had same issue and she reported to police and 2 days later the case was closed saying they can’t do anything.
@gunnarmorling@jaykreps 30-50 will be a large org. I think even things like MVP needs to be questioned, ideal team size and scope of product needs to be questioned, it’s a reset for IT, it will change in different way than anyone can imagine. We need to learn new ways
@klylebell@masterzorgon@claudeai I think you are right, in fact it can lead to explosion of B2B Saas people who can build Saas for targeted audience like 50-500 to make it viable business keep the cost low and take some share from big players 🤔
@gunnarmorling Question requires follow up questions, I will assume we want to go fastest, pipeline it, don’t copy try to get zero copy, pinned CPU, make sure data CPU cache misses are kept minimum. Branch miss prediction is expensive minimise that, is maze how fast it can go
@mwfowlie@gunnarmorling If it’s industry standard the industry is a scam, plain and simple it’s misleading, it’s deliberate to create illusion of huge improvement
@Asandax6@SmitterHane@vibingmonk Different C compilers can generate different machine codes, but in our case we have same bytecode, bytecode is executed by JVM we have different vendors providing JVMs with different performance characteristics, bytecode is no aware what language was used.
@MerrynSW@John_Stepek Nothing wrong with that. Pretty much most decision we take, we think about our own finances. It’s a good indicator for government to understand public sentiment, its government responsibility what’s to decide what’s practical and affordable for country and economy.