Gergely Orosz’s Post

I didn't think code reviews were trivial especially if Claude is writing the code too. Having a form of human control is an essential part of the development process when using AI. Where do they see humans introducing this verification of the AI output?

Gergely Orosz any resource you can recommend that talks about best practices around this?

this is familiar 😇 feels reassuring do remember Dario prediction, and think is was hugely misinterpreted, and probably even he understated the impact. He never said developers weren't in the driver seat. I listened that section several times. Just that would be AI "typing" most code. Intellisense was already writing a lot of code in that sense. I code since basic on spectrum, pascal, c, asp, vbasic, vbnet, c#, sql, tsql, r, python,never stopped. I'm coding 10x these days, but typing is now rare. Reading way more, Learning way faster also, higher focus on concepts, not syntax. Not low level memorization. No grunt work.(mostly) Rulebooks for this new world are still to be written, still too soon.Real challenges, how to scale quality with such amount of code being written, security/safety. but dont feel this is going back.

Gabriel Machado Franco

I watched the full interview and it’s true that he wasn’t saying all software engineering but 90% of “the code” by basically today and “essentially all the code” by end of year. He was also speaking to a general audience that doesn’t have a clear idea of the distinction between SWE and writing code. He said that there are other parts of the job that would be picked away over time, as would most other human jobs. So ultimately he did literally say there will be no need for software engineers, just not on a timeline.

No I think Dario was just straight up lying

I wonder what the quality is, really. Because I love working with Claude but the quality of it is average at best, never superb. Especially tests. Like holy crap it's a showcase of how it learned from humans and how humans are horrible at it...

Would really love to learn more how their teams do it, and I can pick up something new :)

On the review process... That's pretty interesting. But on the testing... I have questions. I'm going to give them the benefit of the doubt that I'm missing something fundamental (because, otherwise, this sounds like bad practice). What is their process for ensuring tests are actually verifying business requirements rather than just reflecting back the implementation that is being tested? On the face if it, auto generated tests are useless because they aren't testing anything other than that the code is doing what Claude thinks it's doing. And you might argue that it's the engineer's job to make sure the test is correct, but, how many times have we *thought* the test is verifying the behaviour when it wasn't (which is the whole philosophy behind TDD, right?). The low barrier to commit of these tests creates a disinsentive to actually verify the tests.... Especially given the culture of "don't ask me to add a test" that your description suggests exists there. Bottom line... I have my doubts this is actually working in practice.

Sounds like a real game changer. However, I have my reservations since it’s not clear to me how useful it is that AI review its own code. Wouldn’t it be the same as if a human author/developer would review its own changes? For obvious reasons, it’s usually a second pair of eyes to conduct the review. Same should apply to AI so I’d assume it implies giving different instructions for coding vs review phase. Really looking forward to the article to get a clear picture of how they do it.