Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

A staff engineer's journey with Claude Code

sanity.io

Post reply on HN
Print view View original

11–20 of 410 posts

Re: A staff engineer's journey with Claude Code

#11

Post by shinecantbeseen »

soperj wrote: post #5
furyofantares wrote: post #2

I've come around on something like this. I start by putting a little effort into a prompt and into providing context, but not a ton - and see where Claude Code gets with it. It might even get what I asked for working in terms of features, but it's garbage code. This is a vibe session, not caring about the code at all, or hardly at all. I notice what worked and what didn't, what was good and what was garbage -- and al…

and do you find this takes longer or shorter than just doing it yourself from scratch?

I’m with you. Sometimes it really just feels like we’re just tacking on the cognitive load of managing the drunk senior in addition to the problem of hand instead of just dealing with the problem at hand.

Re: A staff engineer's journey with Claude Code

#12

Post by rvnx »

swframe2 wrote: post #3

Preventing garbage just requires that you take into account the cognitive limits of the agent. For example ... 1) Don't ask for large / complex change. Ask for a plan but ask it to implement the plan in small steps and ask the model to test each step before starting the next. 2) For really complex steps, ask the model to write code to visualize the problem and solution. 3) If the model fails on a given step, ask it t…

Your tips are perfect.

Most users will just give a vague tasks like: "write a clone of Steam" or "create a rocket" and then they blame Claude Code.

If you want AI to code for you, you have to decompose your problem like a product owner would do. You can get helped by AI as well, but you should have a plan and specifications.

Once your plan is ready, you have to decompose the problem into different modules, then make sure each modules are tested.

The issue is often with the user, not the tool, as they have to learn how to use the tool first.

Re: A staff engineer's journey with Claude Code

#13

Post by furyofantares »

soperj wrote: post #5
furyofantares wrote: post #2

I've come around on something like this. I start by putting a little effort into a prompt and into providing context, but not a ton - and see where Claude Code gets with it. It might even get what I asked for working in terms of features, but it's garbage code. This is a vibe session, not caring about the code at all, or hardly at all. I notice what worked and what didn't, what was good and what was garbage -- and al…

and do you find this takes longer or shorter than just doing it yourself from scratch?

Quite a bit shorter. Plus I can do the a good chunk of the work (first iteration) in contexts where I couldn't before, where I require less focus, and it uses less of my energy.

I think I can also end up with a better result, and having learned more myself. It's just better in a whole host of directions all at once.

I don't end up intimately familiar with the solution however. Which I think is still a major cost.

Re: A staff engineer's journey with Claude Code

#14

Post by MikeTheGreat »

swframe2 wrote: post #3

Preventing garbage just requires that you take into account the cognitive limits of the agent. For example ... 1) Don't ask for large / complex change. Ask for a plan but ask it to implement the plan in small steps and ask the model to test each step before starting the next. 2) For really complex steps, ask the model to write code to visualize the problem and solution. 3) If the model fails on a given step, ask it t…

Genuine question: What do you mean by " ask it to implement the plan in small steps"?

One option is to write "Please implement this change in small steps?" more-or-less exactly

Another option is to figure out the steps and then ask it "Please figure this out in small steps. The first step is to add code to the parser so that it handles the first new XML element I'm interested in, please do this by making the change X, we'll get to Y and Z later"

I'm sure there's other options, too.

Re: A staff engineer's journey with Claude Code

#16

Post by Benjammer »

MikeTheGreat wrote: post #14
swframe2 wrote: post #3

Preventing garbage just requires that you take into account the cognitive limits of the agent. For example ... 1) Don't ask for large / complex change. Ask for a plan but ask it to implement the plan in small steps and ask the model to test each step before starting the next. 2) For really complex steps, ask the model to write code to visualize the problem and solution. 3) If the model fails on a given step, ask it t…

Genuine question: What do you mean by " ask it to implement the plan in small steps"? One option is to write "Please implement this change in small steps?" more-or-less exactly Another option is to figure out the steps and then ask it "Please figure this out in small steps. The first step is to add code to the parser so that it handles the first new XML element I'm interested in, please do this by making the change X…

My method is that I work together with the LLM to figure out the step-by-step plan.

I give an outline of what I want to do, and give some breadcrumbs for any relevant existing files that are related in some way, ask it to figure out context for my change and to write up a summary of the full scope of the change we're making, including an index of file paths to all relevant files with a very concise blurb about what each file does/contains, and then also to produce a step-by-step plan at the end. I generally always have to tell it to NOT think about this like a traditional engineering team plan, this is a senior engineer and LLM code agent working together, think only about technical architecture, otherwise you get "phase 1 (1-2 weeks), phase 2 (2-4 weeks), step a (4-8 hours)" sort of nonsense timelines in your plan. Then I review the steps myself to make sure they are coherent and make sense, and I poke and prod the LLM to fix anything that seems weird, either fixing context or directions or whatever. Then I feed the entire document to another clean context window (or two or three) and ask it to "evaluate this plan for cohesiveness and coherency, tell me if it's ready for engineering or if there's anything underspecified or unclear" and iterate on that like 1-3 times until I run a fresh context window and it says "This plan looks great, it's well crafted, organized, etc...." and doesn't give feedback. Then I go to a fresh context window and tell it "Review the document @MY_PLAN.md thoroughly and begin implementation of step 1, stop after step 1 before doing step 2" and I start working through the steps with it.

Re: A staff engineer's journey with Claude Code

#17

Post by sfjailbird »

shinecantbeseen wrote: post #11
soperj wrote: post #5

Earlier quoted context omitted.

and do you find this takes longer or shorter than just doing it yourself from scratch?

I’m with you. Sometimes it really just feels like we’re just tacking on the cognitive load of managing the drunk senior in addition to the problem of hand instead of just dealing with the problem at hand.

A hundred times more time is spent reading a given piece of code, than it took writing it, in the lifetime of that program.

OK I made up the statistic, but the core idea is true, and it's something that is rarely considered in this debate. At least with code you wrote, you can probably recognize it later when you need to maintain it or just figure out what it does.

Re: A staff engineer's journey with Claude Code

#19

Post by awesome_dude »

axus wrote: post #18

I like his point about more objectivity and zero ego. You don't have to worry about hurting an AI's feelings or your own when you throw away code.

But I still find myself needing (strongly) to let Claude know when it's made a breakthrough that would have been hard work on my own.

Re: A staff engineer's journey with Claude Code

#20

Post by wordofx »

rvnx wrote: post #12
swframe2 wrote: post #3

Preventing garbage just requires that you take into account the cognitive limits of the agent. For example ... 1) Don't ask for large / complex change. Ask for a plan but ask it to implement the plan in small steps and ask the model to test each step before starting the next. 2) For really complex steps, ask the model to write code to visualize the problem and solution. 3) If the model fails on a given step, ask it t…

Your tips are perfect. Most users will just give a vague tasks like: "write a clone of Steam" or "create a rocket" and then they blame Claude Code. If you want AI to code for you, you have to decompose your problem like a product owner would do. You can get helped by AI as well, but you should have a plan and specifications. Once your plan is ready, you have to decompose the problem into different modules, then make…

> Most users will just give a vague tasks like: "write a clone of Steam" or "create a rocket" and then they blame Claude Code.

This seems like half of HN with how much HN hates AI. Those who hate it or say it’s not useful to them seem to be fighting against it and not wanting to learn how to use it. I still haven’t seen good examples of it not working even with obscure languages or proprietary stuff.

Post reply on HN