This article is somewhat reassuring to me, someone experimenting with openclaw on a Max subscription. But idk anything about the blog so would love to hear thoughts.
In my opinion (which means nothing). If you are using your own hardware and not profiting directly from Claude’s use (as in building a service powered by your subscription). I don’t see how this is a problem. I am by no means blowing through my usage (usually <50% weekly with max x5).
I think this is better than many current coding interview methods. Assuming you have an agent setup to not give the interviewee the answer directly.
Of course there are times when you need someone extremely skilled at a particular language. But from my experience I would MUCH prefer to see how someone builds out a problem in natural language and have guarantees to its success. I’ve been in too many interviews where candidates trip over syntax, pick the wrong language, or are just not good at memorization and don’t want to look dumb looking things up. I usually prefer paired programming interviews where I cater my assistance to expectations of the position. AI can essentially do that for us.
Yeah research says the interview process should match the day to day expectations as closely as possible, even to a trial day/week/month. All these leetcode and tricky puzzles are very low on signal. They don't tell you how a person will do on the job at all, not to mention they're bad for women and minorities.
This type of coding has been extremely helpful to me in the past few weeks. I’m on parental leave, but also a co-owner of a small company and can’t completely log off.
I can one handed spec out changes, AI does its thing, and then I review and refine it whenever my kid is asleep for 20 minutes. Or if I’m super tired I’m able to explain changes with horrible english and get results. At the same time, I am following a source control and code review process that I’ve used in large teams. I’ve even been leaving comments on PRs where AI contributes and I’m the only dev in the codebase.
I wouldn’t call this vibe coding— however vibe coding could be a subset of this type of work. I think async coding is a good description, but bad because of what it means as a software concept (which is mentioned). Maybe AI-delegation?
AI has drastically changed how I make decisions about code and how I code in general. I get less bogged down with boilerplate code and issues, which makes me more efficient and allows me to enjoy architecting more. Additionally, I have found it extremely helpful in writing lower-level code from scratch rather than relying on plug-and-play libraries with questionable support. For example, why use a SQLite abstraction library when I can use LLMs to interact directly with the C source code? Sure it’s more lines of code, but I control everything. I wouldn’t have had the time before. This has also been extremely helpful in embedded systems and low-level Bluetooth.
In terms of hiring- I co-own a small consultancy. I just hired a sub to help me while on parental leave with some UI work. AI isn’t going to help my team integrate, deploy, or make informed decision while I’m out.
Side note, with a newborn (sleeping on me at this moment), I can make real meaningful edits to my codebase pretty much on my phone. Then review, test, integrate when I have the time. It’s amazing, but I still feel you have to know what you are doing, and I am selective on what tasks, and how to split them up. I also throw away a lot of generated code, same as I throw away a lot of my first iterations, it’s all part of the process.
I think saying “AI is going X% of my work” is the wrong attitude. I’m still doing work when I use AI, it’s just different. That statement kind of assumes you are blindly shipping robot code, which sounds horrible and zero fun.
> why use a SQLite abstraction library when I can use LLMs to interact directly with the C source code?
Because of the accumulated knowledge in these abstraction layers and because of the abstraction itself resulting in readable and maintainable code.
Yes you can move the abstraction one level up, but you don't control it if you nor the LLM meet the level of accumulated knowledge that is embedded in this abstraction. Let alone future contributors to your codebase.
Of course it is all depending on context and there is no one-fits-all strategy here.
awesome article. Do you know how to access the other articles in the series that are linked (dead) in the first paragraph? Would be great to read if still available.
N=1 is addressed, see outcome predictions. N=1 comes with caveats, of course, but a study like this, with a proven harmless supplement, should be welcomed and praised.
It is clearly a step forward from what you can watch about theanine on YouTube or TikTok. I consider this a work of citizen science. While it should not be taken for more than it is, it’s a great example of how someone can experiment without a high burden.
N=1 studies aren’t evil. They’re just pretty close to the entire history of pre-modern medicine that led us to bad evidence. My concern here is not that someone is sharing their opinions, it’s the fact that the person doing this explicitly heaps derision on the “placebo people” (or some other phrasing) and then heaps praise on other people doing N=1 studies and proceeds to do one. This stuff all needs to be treated with humor, good faith, and then extreme skepticism about any result it produces.
That's a pretty low bar though. OK, it's one step up from a monetized YouTube video that boils down to "It works--Trust me, bro." I still wouldn't really call it citizen science.
All of raspberry pi’s SDK documentation and data sheets for the pico (rp2040) are amazing. Perfect material to build an embedded engineering course around.
https://thenewstack.io/anthropic-agent-sdk-confusion/
In my opinion (which means nothing). If you are using your own hardware and not profiting directly from Claude’s use (as in building a service powered by your subscription). I don’t see how this is a problem. I am by no means blowing through my usage (usually <50% weekly with max x5).
reply