Hacker News Reader

Comment by DaSHacka | Hacker News Reader

I honestly can't imagine using a bigger plan. The $20 a month plan is able to be used almost all day without hitting the usage quota so long as you moderate what model you're using.

I really don't understand why so many people seemingly use the bigger plans. Are people just running opus 5.5 as 4+ subagents 24/7?

In order for me to actually understand what's being done, I always keep it brief and use specific prompts (and usually use smaller models since they're perfect for that). Do people just let "claude take the wheel"?

When you have to go in to your next standup and explain what your code does to your boss, and what you've done since the last meeting, what do you say?

View on HN

I let Claude take the wheel with various non-professional projects which often means heavy subagent workflows. Prompts like this easily eat through my 5x Max plan limits:

> I want you to look for UI issues related to transitions. Look for jumps, reverts, or anything that doesn't follow UI/UX best practices. You might need to figure out how best to even record and review UI transitions. Do all testing locally on a local dev server, but make sure to undo any database changes you make in the process of testing. Be exhaustive and report back with an .html report that highlights the visual issues and proposed solutions so I can review each one. Ultracode on.

Which means I have up to 10 Opus 5.5 medium agents running concurrently for long stretches of time.

I want to be clear I don't think that using Claude in this way gets you the best possible results. But when the app is fully vibecoded then audit prompts like the one above can be seen as a definite net positive. Making Claude think really hard about various improvement routines will get you better results in that one aspect than Claude's first attempt to build something from scratch did.

> When you have to go in to your next standup and explain what your code does to your boss, and what you've done since the last meeting, what do you say?

You say the same thing to your boss as if you were a lead or manager reporting on a project that your junior and mid-level devs are working on. "What" is the easy part to explain; "how" is the details that your manager doesn't need to hear about if you're doing your job properly.

I built an MCP server for Altium that lets us do automated PCB design review and AI assisted PCB design, that cost a good $1500 in tokens.

I experimented with using Astra and the OnShape API to take one of our robot component designs and make a new version of a different size by finding components from the same families of a different size and building a new scaled design against it. That cost about $120. I continued with that and explored two additional design variants. That session cost $2000. This work is a capabilities check not production intent hardware, but I need to be able to explain to the mechanical engineers what is possible.

I took a robot simulator that runs in Gazebo and had it rewritten in rust and WASM to run on any machine in the browser.

At home I’ve been building an AI first PCB design suite along with a fast autorouter. There is a corpus of 1300 PCBs I have it route. I am building a mechanical CAD system as well, and am working on combining the PCB design program and the mechanical CAD program in to one MIT licensed electromechanical design suite you can use with AI tools to build complete products. I am using AI to design a POE powered studio monitor speaker, and I’m doing a lot of work with LED pendant designs. I have a few video game projects I am working on. I have two MAX 20x plans and I run them both out of credits every week. Again this part is at home, the earlier examples were at work.

When agentic coding popped off in January, I wanted to see what the new world would be like. I’ve become very comfortable with these tools and one thing I’ve learned is that when the tool only needs you once every 15 minutes, it’s helpful to have multiple things going at once.

> When you have to go in to your next standup and explain what your code does to your boss, and what you've done since the last meeting, what do you say?

I hand them a writeup of how incorporating agentic decision making into cleaning up some of our messy data saved us $XXX,XXX at a cost of $XX of tokens, with the receipts attached.

> Do people just let "claude take the wheel"?

Yes, I'd wager most people just let "claude take the wheel".

> When you have to go in to your next standup and explain what your code does to your boss, and what you've done since the last meeting, what do you say?

You say "Claude, I've got a standup in 10 min, give me a high level summary for my boss of everything you did since the last standup" and then you read that lol.

I run 4 agents under one supervisor agent for 12 hours a day building computer use environments. Yes, I don't read the code, but testing is 90% of my work. Got to find other ways to buy trust in AI code. I came up with a stack of 5 layers of testing which is pretty similar to the recent Anthropic writeup. I can consume the $100/mo plan weekly quota in 2 days.

Sounds cool. Can you link me to that write up? Appreciate it.

How do you get along with the 20USD plan and that much utilization?

I use Claude mainly via Webinterface and when on Opus 4.8 or higher, I reach my weakly limit around 2 - 3 days before rest - and Im just prompting single/specific very limited & constrained classes & functions with proper description and architecture upfront.

The Claude Code CLI is far more efficient at many tasks than the web interface or the desktop app.

Example of token-heavy jobs:

- PCB layout. Last job took 8 hours unsupervised.

- CAD design. Less heavy, but still heavy.

Coding is lightweight for an LLM, depending on how complicated testing is.

Yep. I’m building a PCB autorouter [1] and testing it on a corpus of 1300 boards. I’m using so much conventional compute just running the test suite I had to stand up an old desktop and build a reservation queue hah. But visually debugging auto routing failures on a big corpus chews through usage, among other things I’m doing.

[1] https://github.com/sequoia-hope/ultraroute See also https://github.com/sequoia-hope/boardroom

My guess is that autorouters can still make sense for bulk routes (memory busses?). But for routing you'd otherwise do by hand, I'd prefer an LLM. That would be needed for component placement and zones anyway right? I guess the autorouter is still essential, for these massive busses, which you just cannot do with an LLM. But I've never done such boards.

My system works with both. An LLM can hand route but it can also use the autorouter when appropriate. One goal of my system is to build a detailed database of common circuits and allow in-browser design and routing of basic boards using all classical algorithms without the need for any LLM. The pitch is “build circuit boards like LEGO in a browser tab”. The desktop version also has full LLM integration for more complex design work.

But basically it’s an AI-native EDA suite. It’s the tool I want for my professional work that I can also hand to my friends who don’t know PCB design, flexible enough to do both.

The real question is whether the work will still be valuable in six months (the bitter lesson) but for now I gotta build to understand the future.

E.g. like a template for laying out a regulator with support components around it, or e.g. a USB hub with it's reference circuit laid out around it.

Reply on news.ycombinator.com