Hacker News Reader

Comment by retrac | Hacker News Reader

I have had some truly spectacular results that still kind of stagger me in the last few months using Claude in my hobby projects -- but even though Claude insists on trying to slip its name into the git history as credit it's not Claude -- it's me. Someone who has studied CS and software engineering for decades will craft different prompts from someone without that background. A suggested axiom: there is nothing I can build with Claude that I could not build myself with my current level of CS knowledge, assuming I had infinite focus and time. In my hands it can go as far I could anyway, and no further. (But it is faster!) My experience bears that out so far.

View on HN

Fair enough but speed, especially the kind that comes with LLMs, is fast enough to open new ways of working and doing things. We don't have infinite time and if there's something that can give me multiple, for example, UI suggestions in a minute which I can pick from, it's a different way of working than sitting with a UI designer for several hours have discussions. So, while I agree with you in theory, I don't fully agree with you in, what I think you're implying, when it comes to practice.

> hobby projects

Unfortunately despite being impressive for solo stuff, such results don’t scale to software you’d give to others.

Claude writes probably 95% of our code now, fintech, amongst top 5 in the world in what we do. I am 100% certain we're not even at the forefront of using agents for coding compared to some others.

It definitely can scale.

Are there any code reviews? Or is AI reviewing code also?

We still review everything. And we guide by planning, prompting, speccing.

So we're not actually much faster at the core code, because reviews still take time. Ultimately, we're on the finance markets and we have regulatory pressures and I, as the human, am responsible for putting the code out there.

But we're freeing up a lot of time to get other things correct. We have n x more metrics now because plumbing in basic stuff is now trivial. We now have dozens more tools and skills to help analyse issues (e.g. why this price and not x), answer questions, etc.

I now have skills to scrape logs, download unpack and scape our bus persistence, link to kdb, and so on, all in my claude prompt, joining it all together and the AI is figuring things out. I can diagnose things, I don't know, maybe 100 times faster?

It is revolutionary, and I am highly sceptical of the motivations of people who keep saying otherwise.

When you review everything, do you understand every line of code before approving? Do you make it rewrite code that is too abstract or unclear for future humans to understand? Does AI write the tests and do you review those with the same diligence?

I don't disagree that it's revolutionary in many ways, but I am seeing lots of companies make very costly mistakes by relying too heavily on AI without fully understanding the code it writes and without fact checking its outputs by a human.

Yes we understand all code merged. Yes, I have coding rules and standards that the agent follows to ensure it writes code in virtually the same style as the whole team does. Someone isn't going to let my change be merged if they don't understand it.

My hunch however is that in a few years we're not going to be reviewing as heavily as we improve guardrails and can trust the AI code and review cycle more. I'm not sure what to expect for my career at that point.

> Someone who has studied CS and software engineering for decades will craft different prompts from someone without that background.

This, to me, is the biggest differentiator. In terms of results, there's a huge yawning chasm between the person who says "Claude make me a $thing" versus the person who puts in the effort to lay down the overall architecture, gives some thoughts to libraries and dependencies, performance trade-offs etc, and only then begins prompting.

Knowing how to implement Djikstra or a linked list by heart is no longer important. Actual software engineering skills are more important than ever.

> Knowing how to implement Djikstra or a linked list by heart is no longer important.

This was never important. The important part was always knowing when to use them.

>The important part was always knowing when to use them.

Two things can be true simultaneously. I think there was a time when deep familiarity with implementing algorithms was important.

always was. Still is.

It is for coding interviews

Even more narrow - you only need to know when to consider them.

I only ever bothered remembering enough of any algorithm to know my options and a few rules of thumb. If I ever actually need to consider the details of the algorithm, I certainly need to spend a lot more time thinking through the problem and its solution. Knowing a specific algorithm well enough to pump out in 15 minutes is a party trick that is as useful as being able to change a tire in 3 minutes flat. A great time saver that will be functionally useful maybe 3 times in your life...

The gap is closing; a shitty wannabe programmer will eventually learn the structures one way or another. Agentic coding just got many new people involved, and these new people create noise.

Reply on news.ycombinator.com