brunooliv

Couldn’t read the entire comments but, my experience has been overwhelmingly positive so far. I think what helps me be effective is a combination of factors: I work only in a modern, well-documented and well-architected Java codebase with over 80% test coverage.

I only use Claude Code with Opus 4.6 on High Effort.

I always, ALWAYS treat my “new job” as writing a detailed ticket for whatever it is I need to do.

I give the model access to a DB replica of my prod DB that I create manually.

I do NOT waste time with custom agents, Claude.md files or any of that stuff.

When I put ALL of the above together, the results ARE THE PROMISED LAND: I simply haven’t written a single line of code manually in the last 3 months.

What rock?

C'mon let's be real here, there's either "testing AI skills" versus "using AI agents like you would on the daily".

The signal got from leetcode is already dubious to assert profeciency and it's mostly used as a filter for "Are you willing to cram useless knowledge and write code under pressure to get the job?" just like system design is. You won't be doing any system design for "scale" anywhere in any big tech because you have architects for that nor do you need to "know" anything, it's mostly gatekeeping but the truth is, LLMs democratized both leetcode and system design anyway. Anyone with the right prompting skills can now get to an output that's good for 99% of the cases and the other 1% are reserved for architecs/staff engineers to "design" for you.

The crux of the matter is, companies do not want to shift how they approach interviews for the new era because we have collectively believed that the current process is good enough as-is. Again, I'd argue this is questionable given how sometimes these services break with every new product launch or "under load" (where YO SYSTEM DESIGN SKILLZ AT).

I really love Anthropic's models, but, every single product/feature I've used other than the Claude Code CLI has been terrible... The CLI just "sticked" for me and I've never needed (or arguably looked in depth) any other features. This for my professional dayjob.

For personal use, where I have a Pro subscription and adventure into exploring all the other features/products they have... I mean, the experience outside of Claude Code and the terminal has been... bad.

Would be really curious to see

Mickens is the best!

It should be fixed now!

Yes it assumes a minimum running base, will improve thx

I built the running app I always wanted: https://runcoach.fly.dev

You get tailored running schedules and also some body weight strength workouts and healthy meals all in one!

Even if it broke after some sort of vibe coding session, the fact that we’re now pushing these tools to their limits are what’s allowing Anthropic and Boris getting a lot of useful insights to improve the models and experience further! So yeah, buckle up, bumps expected

Sorry for the hiccup, it should be better now!

Over the holiday period, I have vibecoded a simple running/fitness app that I wish I had, but never found!

The app is geared towards running, offering comprehensive plans for different distances and durations, with the ability to log runs/save plans, export to PDF and focused on simplicity.

It uses a progressive distance building approach to allow runners to safely reach their goals.

As an extra, it offers some dedicated strength training for running as well as healthy/hearty meals focused on lean protein, veggies and all the cliches :)

I know there's nothing new or unique about an app like this, but, I like that I have one single place where I can generate a running plan, have some nice meal ideas for when I lack inspiration, as well as some strength training added in the mix!

If this resonates and it's helpful, there's an option to buy me a coffee which will help cover all the vibecoding/hosting costs :)

I hope it's as useful for someone as it has been for me (currently following a plan generated with it for a trail run in a few months' time!).

Feedback welcome!

I agree completely, I meant in terms of opensource ones only. Opus 4.5 is the current SOTA and using it in Claude Code is an absolute amazing experience. But, paying 0 to test GLM-4.7 with opencode, feels like an amazing deal! I don’t use it for work though. But to keep “gaining experience” with these agents and tools, it’s by far the best option out there from all I’ve tried.

GLM-4.7 in opencode is the only opensource one that comes close in my experience and probably they did use some Claude data as I see the occasional You’re absolutely right in there

Inspired by a lot of back-and-forth work with several different agents, as well as shaped by my own understanding of using context windows (and tokens) in as efficient a way as possible, the LLMs and myself have came up with a quite cool system heavily inspired by Anthropic's skills to dynamically extend an agent's capabilities. Based on how it works, the name "Lazy Skills" was the most logical one!

Read for my attempt to answer this question!

The thing is all these big labs are so “transformer-pilled”, and they need to keep the money furnaces growing that I think it’ll take considerably more than 10 years, more like 20-30 if we’re lucky.

I both agree and disagree with this post, but I might be misunderstanding it. Near the end, it states:

“Enjoy writing it, it doesn’t have to be nice or pretty if it’s for you. Have fun, try out that new runtime or language.”

It doesn’t have to be nice or pretty EVEN if it’s NOT for you. The value in prototyping has always been there and it’s been very concrete: to refine mental models, validate assumptions, uncover gaps in your own thinking (or your team’s), you name it.

Unfortunately it feels that the pendulum has swung in the completely opposite direction. There’s a lot of “theatre” in planning, writing endless tickets and refining them for WEEKS before actually starting to write code, in a way that’s actively harmful for building software. When you get stuck in planning mode you let wrong assumptions grow and get baked in into the design so the sunken cost keeps rising.

Simply have a BASIC and SHARED mental model of the end goal with your team and start prototyping. LLMs have made this RIDICULOUSLY CHEAP. But, the industry is still stuck in all the wrong ways.

I’ve been reading the comments here… idk which companies you all work at, but, at least in all places I’ve worked you’ll have 2/3 people who are REALLY GOOD. Then you’ll have like 3/4 people who can sort of “trail” after those. And then there’s “the rest”. I’ve seen “the rest” making informed and “well planned” decisions that are just terrible approaches to a given problem. An LLM can actually be a net positive versus this rest.

As a non-UI/UX person:

“Synthesizes learnings into a scalable and adaptive layout.”

Genuinely curious: what is inherently scalable and adaptive in a three panel layout?

Agreed! It’s been horrible recently, feels like a completely different model under the hood. Before I could use it as a real sparring partner for architecture designs and decisions and I actually would learn in the process. Now it’s like it’s sycophancy is tuned to the max, it just agrees with me, does the bare minimum and produces code that doesn’t compile. For that I have the humans, ah!

Nice to see!! I wrote a longer post about this: https://open.substack.com/pub/boliv/p/claude-code-usage-patt...

Am I the only one who uses CC in a one-shot fashion? I rely on what has been generated by the init slash command at the root of my monorepo and then when I have a complex task I manually create a md file called CONTEXT.md inside the package/area where I want to implement the new task. I try to be as detailed as possible, think like writing a ticket for a junior dev. Then my only command for CC is something like: “Focus on the project structure described in CLAUDE.md and work through the task planned in CONTEXT.md. Follow current codebase standards, keep code reusable and easy to follow and modify.”

This is like 99% of my CC interactions working on top of a well structured codebase and it just works perfectly for almost any task I throw at it.

It’s a thin line to walk for me, but I feel that the whole “skill atrophy” aspect of it is the hardest to not slip into. What I’ve personally liked about these tools is that they give me ample room to explore and experiment with different approaches to a particular problem because then translating a valid one into “the official implementation” is very easy.

I’m a guy who likes to DO to validate assumptions: if there’s some task about how something should be written concurrently to be efficient and then we need some post processing to combine the results, etc, etc, well, before Claude Code, I’d write a scrappy prototype (think like a single MVC “slice” of all the distinct layers but all in a single Java file) to experiment, validate assumptions and uncover the unknown unknowns.

It’s how I approach programming and always will. I think writing a spec as an issue or ticket about something without getting your hands dirty will always be incomplete and at odds with reality. So I write, prototype and build.

With a “validated experiment” I’d still need a lot of cleaning up and post processing in a way to make it production ready. Now it’s a prompt! The learning is still the process of figuring things out and validating assumptions. But the “translation to formal code” part is basically solved.

Obviously, it’s also a great unblocking mechanism when I’m stuck on something be it a complex query or me FEELING an abstraction is wrong but not seeing a good one etc.

I think a big reason why Claude Code is winning is because it’s such a thin wrapper over a very strong base model which is why people are afraid of comparing it directly. All these IDE integrations and GUIs and super complex system prompts etc are only bloating all these other solutions with extra complexity, so comparing something inherently different becomes also harder.

Damn, this is the type of code that probably makes sense to the original writer but good luck maintaining it. Even AI would struggle with this type of convoluted conditional logic without a test harness. None of these things should be merged without tests. Even basic unit tests would catch these. Code quality is an illusion.

Agreed, this is just terrible for the field as a whole it’s like we’re regressing or something

I can totally relate to that feeling but after a few rejections, and having to work 9-5 during the day and prepare in the evenings with a drained battery, it just totally kills the motivation. Maybe it's the type of circumstance when if I'd really need I could do it, but to try to "suffer" through the process just to be rejected is quite hard.

In my experience this never worked for me for the simple fact that I never had the skill and time to even GET a second offer to bargain against. I wouldn’t say I’m a bad developer, but I’m definitely horrible at interviewing and having to jump through several different steps at different companies just to bargain for a higher end of the range makes me feel like either the market is the problem or the expectations from companies are way unrealistic

Claude 3.7 Sonnet nails this:

> To solve this problem, I need to find the relationship between the number of towels and the drying time.

Given information: - 2 towels take 2 hours to dry

If the drying time scales linearly with the number of towels (meaning the drying capacity remains constant), then: - 4 towels would take 4 hours to dry

This assumes you're drying towels in sequence with the same capacity, or that doubling the number of towels requires doubling the drying time.

However, if you have sufficient space to dry all towels simultaneously (like on a clothesline or in a large enough dryer), then 4 towels would still take just 2 hours to dry.

Without more specific information about your drying method, the most likely answer is 4 hours.

draw.io