- at first LLMs apply a penalty to development: it requires to explore new toolings for AI assisted coding, then select and get used to one of them, that might well end up being a transient solution. e.g. Cursor, to be replaced soon by Claude Code?
- new uncertainty about which feature to develop: if I can develop it, then anyone else surely can reverse engineer it easily and replicate it? This is quite unnerving.
The reason is that the post you link to is overly simplistic. The only reason why Simon's experiment works is because there is a pre-existing language agnostic testing framework of 9000 tests that the agent can hold itself accountable to. Additionally, there is a pre-existing API design that it can reuse/reappropriate.
These two preconditions don't generally apply to software projects. Most of the time there are vague, underspecified, frequently changing requirements, no test suite, and no API design.
If all projects came with 9000 pre-existing tests and fleshed-out API, then sure, the article you linked to could be correct. But that's not really the case.
If you start with some working software, you could make an LLM generate a lot of tests for the existing functionality and ensure they pass against the existing software and have excellent test coverage. Generating tests and specifications from existing software is relatively easy. It's very tedious to do manually but LLMs excel at that type of job.
Once you have that, you port over the tests to a new language and generate an implementation that passes all those tests. You might want to do some reviews of the tests but it's a good approach. It will likely result in bug for bug compatible software.
Where it gets interesting is figuring out what to do with all the bugs you might find along the way.
> pre-existing language agnostic testing framework of 9000 tests
if there exists a language specific test harness, you can ask the LLMs to port it before porting the project itself.
if it doesn't, you can ask the LLM to build one first, for the original project, according to specs.
if there are no specs, you can ask the LLM to write the specs according to the available docs.
if there are no docs, you can ask the LLM to write them.
if all the above sounds ridiculous, I agree. it's also effective - go try it.
(if there is no source, you can attempt to decompile the binaries. this is hard, but LLMs can use ghidra, too. this is probably unreasonable and ineffective today, though.)
Couldn't find it on the Play store by searching for the name and the developer's name: if it is not just me then your app is very hard to discover.
So I am installing it through the link you provided, which directed me to a "install success" page saying "your purchase is successful" even if your app is free. Another obstacle to adoption :-)
Last, I was not informed on the page of the app' size. Seeing what it does and the time it takes to download I am afraid it could be huge? Third obstacle :-)
Thank you for the feedback, I really do appreciate you taking the time to check it out and write out the comment! I'll look at adding a note about total app size in the description, it won't hurt.
As for discoverability / the "your purchase is successful" message, I'm not sure what else I can do, I've set it to free, no ads etc in Google Play. Maybe I need to hit a few more keywords for transcription so it surfaces it more.
For me, searching for "whistle" on play store, I get the app as the third result (ignoring sponsored crap). Searching for "blazingbanana" gets me the app as the first result".
App info shows 218MB size, which I suppose is about what I'd expect for a model+app code :shrug:
Good to know, it's hard to know what real users would see in the play store and not Google just showing you what you want. Thank you for checking it out
I hesitate trying Claude Code, there is a bit of fatigue installing new tools :-)
By the way: ChatGPT 5.1 Codex Max pro, which was rubbish when used directly inside Cursor, is actually very good once I access it via the OpenAI Codex extension installed within Cursor!
Mysteries of toolchains...
Anyway, that proved so useful when I maxed out my Claude credits.
To nick pick, it is more about 800% gain in productivity. Cost actually increased.
I refer here to my experience as a solo developer.
With AI assistance I don't spend less hours coding, but more.
There is the thrill of shipping relevant features that were sleeping in my drawers for ages, quicker. Each hour of coding delivers just 8 x more features and bug solving.
Also, whereas I spent a few dozen dollars per month on server costs, I now also spend an equivalent amount on subscriptions and API calls to LLM services for this AI assisted coding. Worth every penny.
So while productivity increased manifold, absolute cost actually increased as well for me.
- traffic is slightly but noticeably decreasing for the last year: https://public.nocodefunctions.com
- it pushes me to improve the design, the overall quality, explore new feature spaces: https://next.nocodefunctions.com
Also, I noticed:
- at first LLMs apply a penalty to development: it requires to explore new toolings for AI assisted coding, then select and get used to one of them, that might well end up being a transient solution. e.g. Cursor, to be replaced soon by Claude Code?
- new uncertainty about which feature to develop: if I can develop it, then anyone else surely can reverse engineer it easily and replicate it? This is quite unnerving.