Nostr Web Client

Why is it finally ready now after ten years of being a barely functional input-only android app?

A few weeks ago I saw nostr:nprofile1qqsr7acdvhf6we9fch94qwhpy0nza36e3tgrtkpku25ppuu80f69kfq9q9kky giving a talk and demo of [Shakespear](https://shakespeare.diy/), a Chrome app for vibe-coding.

Explain the app you want, and the model you select will build it. Don't even need to be a dev they reckon.

So I figured I'd give it a try.

Start again from scratch, import the old data.

In about a week of work this app has progressed far beyond the prototypes that spent more then

ten years as half-running shoddy input-only systems that I couldn't be arsed to expand further.

It went [pretty well](https://exocortexlog.com/news/articles/2025-11-25-vibecoding/) to start with,

something even a non-dev could do, then [ceased up](https://exocortexlog.com/news/articles/2025-12-01-fixingvibes/), unable to really understand the codebase it'd written until I spend a fairly long day manually cleaning up it's mess.

So Shakespeare (and presumably all the other tools I haven't tried) seems okay for a non-dev to prototype a small app but currently the models are writing code so sloppy that they can't then later understand it themselves. Still needing a dev's guiding hand to keep it from

repeating itself or creating complex unorganized unmanageable code.

#vibeCoding #shakespeare

nostr:nevent1qqsdr2ww059h02mffsg40g8hdg53e8hxs04m93pgn3qu366yl7syp4cppemhxue69uhkummn9ekx7mp0qgsqmt3djh6a9zguy7466d42fa3dqjmw8lh87m25sf94sjrsxym3jdsrqsqqqqqpvcupkr

Reply to this note

Please Login to reply.

Discussion

Sebastix 0mo ago

https://leadershiplighthouse.substack.com/p/i-went-all-in-on-ai-the-mit-study

Pre 0mo ago

Yeah. A lot of the investment and excitement for the AI coding is based on the faith that it'll get better. Some people expect exponentially so even! Though it looks to me more like diminishing returns on exponentially growing investment really. And to assume it can improve without limits is quite a leap of faith.

Nice too though. So long as you're managing the code itself and not just the UI seems useful.

hasky 0mo ago

Yeah all in .

Derek Ross 0mo ago

I'd say it depends on what you're building. Some applications are more complex than others and AI does well depending on the overall complexit. That said, are you using Claude Opus 4.5? It's expensive, but it's a pure win in my opinion.

Pre 0mo ago

Paid in sats for your basic model for most of it. Realized I was spending too many sats and switched to the OpenAI account. GPT 4.1 really made a mess of things but 5.1 was certainly much better. 4.1 would just delete entire chunks of functions and replace them with comments saying not to change this part 😄

Ought to be possible to improve the models you'd think, but not without limit. Wonder what those limits will turn out to be.

Pre 0mo ago

Some things that I'd like in Shakespeare in the first blog post there [ https://exocortexlog.com/news/articles/2025-11-25-vibecoding/ ].

Mostly better management of the chat context. Able to pull back old ones, and delete chunks from them where the robot goes astray or is too verbose and fill up the context window.

Derek Ross 0mo ago

Okay! Thanks for th feedback. I'll take a look.