Nick Dobos reports staying hands-off from coding for months.
I don't write code anymore
I haven't written code in I think 6 months?
I think everyone is like this no?
@levelsio @TermiusHQ so do you write code? or you just write to Claude what do you want? ex. you want a new feature or you want a change. you just prompt or what ? or Claude is mainly the assistant here? my question mostly is how much trust you put on claude while you are on prod.
Positive users celebrate Levelsio using Claude for all development since it renders manual coding unnecessary for most apps, while negative users reject the approach as useless for proprietary systems and logic tasks.
No Digg Deeper questions have been answered for this story yet.
@levelsio do you still enjoy this kind of building though?
to easy to achieve something. before AI I was at least getting some dopamine rush when something I built worked
@levelsio Did you LLM code all these.
If so. Impressive.
@levelsio I work with proprietary banking software built on top of an Oracle database. Closed source, closed documentation, no community, custom IDE, custom source repo. LLMs can’t do shit and don’t know shit about it. They hallucinate everything. I do not use LLMs as a programmer.
@BitcoinArchive I saved in memory to always test first like php - t file
@levelsio Can you outsource code review to another llm or even an agent and get a reliable check?
@levelsio same but i also never wrote code in the first place
@levelsio i still read/review a lot, but not writing any code by hand
@levelsio I do FPGA very low level, still do it as necessary but the level has dropped significantly. At the same time the productivity has gone up.
@levelsio Depends of the domain
SaaS + CRUD? Pretty much automated by now
AAA Gamedev? Getting tricky because of scale
New foundations? Remains a human holy grail
@levelsio Yup been hands off for months now.
Though, we are still very early and lots of people have still barely touched AI dev.
Just look at how big companies have AI leaderboards because they needed to convince employees to use it.
I don't write code anymore
I haven't written code in I think 6 months?
I think everyone is like this no?
@ffaebi @levelsio You REALLY do not understand. Everything is in the corporate workspace. You can’t install shit, you can’t train shit. All source code is proprietary, every data transfer is monitored and blocked.
@itzsam_ai I see what it edits yes
@levelsio It’s more code reviewing than code writing
It is hard to explain if you dont know the stack. It is really funny what our job looks like, compared to the people who use python + VS Code + Github + idk what (talking out of my ass here) and shipping iOS apps or web apps.
The platform is closed source, build on top of an oracle database. They allow their customers to “parameterize / customize” it. As a result, I’m kind of forbidden from making direct interactions with the database. Everything goes through an interface.
Typical tasks are things like: add a new input form, create a new workflow, design a new document to print and fill it with customer data (which requires deep knowledge of how the bank setup its static data, with all the quirks and exceptions.
The LLM would have to learn the entire kernel codebase, the entire documentation, the entire customer codebase, the entire database structure, the entire Jira history with all the issues, the Confluence specifications. Maybe then it would be able to come up with something.
But don’t LLMs program iteratively? They compile and test? I don’t know how that would work with my software. The installation takes a long time and is triggered from a drag-and-drop tool, no command line available.
@BitcoinArchive @levelsio absolutely, and not only code review but also all the testing phase, you can run linters, sanitizers, unit/stress/fuzz tests and also drive automatic GUI taps and test every use case
same for what comes before even a line of code is written by AI
plan with clear goals & stages
You argue LLMs don‘t know shit which means you haven't tried or just lie. You can also run qwen3.6 fully locally. Big corps have self hosted inference for such reasons, or at least proper contracts. It's just GPUs which get you easily compliant.
Maybe you should say, I work like a monkey, I can't do an anything anyway, but then why bother commenting about LLMs.
@levelsio Coding only makes sense for low level stuff, like extending or building technology.
For higher level stuff such as building apps, not needed anymore.
@levelsio Yup, I wrote code for ~25 years, now I'm in the exact same place that you are, and producing more code, at a higher quality level, than ever before.
@BitcoinArchive @levelsio i usually do a ping pong between chatgpt pro research (1h thinking each iteration) and then let it confute/review by opus 4.7 (max thinking), and for the first couple of days i let them argue about what's the absolute best plan, SOTA model, best agentic scaffolding etc (1/n)
@nickkirt @levelsio We do seem to be stuck in time. Work looks the same as it did 15 years ago. Still writing the simplest functions by hand.