They did it first because doing it first was easier. There are tons of examples around and code can be verified to work.

It would be better for you if they started a new one when your current one expired so that there's less time until you get your new allowance when you happen to start using it.

They might do this for UX reasons (the auto-restart looks weird), or they could be trying to get people to upgrade. Maybe both.

A solution I've found is to just say "yo" to Claude every 5 hours. This somewhat decreases the chance that a deep Claude session gets blocked.

yo-claude is an easy way to enable this. Feedback and bug reports welcome.

Comments URL: https://news.ycombinator.com/item?id=46811648

Points: 1

# Comments: 0

In my own AI-heavy workflows, I’ve noticed that AI tools are great at generating layouts quickly, but the results tend to converge on a certain look and feel and often lack polish around responsiveness and design details.

Templates still accelerate my builds. They encode decisions, constraints, and taste that I don’t want to recreate from scratch... even with the help of a coding agent.

As AI becomes more central to how we build things, do templates continue to retain value, or is this just a transitional phase?

Comments URL: https://news.ycombinator.com/item?id=46546450

Points: 1

# Comments: 1

I built this over the holidays to get myself out of the house more in 2026.

Coffee Hop helps remote workers find cafés (and pubs) nearby, rated by the community for WiFi speed, plug availability, coffee quality, and ambiance.

Pick how far you want to walk (5 mins to an hour), work for a couple hours, then hop to your next spot.

The idea: $10-15/day in coffee vs $50 for a desk in a co-working space, plus you get some movement between sessions.

Built entirely with AI (Cursor and Opus 4.5) - I didn't write any code myself.

Not many ratings yet, but I'm using it myself to build up the initial data.

Comments URL: https://news.ycombinator.com/item?id=46438079

Points: 2

# Comments: 0

https://spikelog.com

https://runnem.com

https://leveloh.com

https://thefudgesisters.com

https://hop.coffee

I'm write a few articles here about tricks that work for me when it comes to AI assisted coding: https://foundinglean.substack.com

I had had the idea and the domain registered for years and recently just took the leap to put it out there.

https://spikelog.com

I have no idea how many people bounced at that step but I know I personally close tabs when something wants my email before I can even look around. So I finally added a "try without signing up" flow. This was recommended by a commenter the original post: https://news.ycombinator.com/item?id=46085379

To get this working when you press 'Try it now', I create a guest user and give you back a 'refresh' secret. That goes in localStorage. Next time you visit, we swap it for a fresh JWT. If you eventually sign up for real, your stuff transfers over.

I'm using a separate (from my auth router) JWT keypair for guests vs real users. Idea being if someone compromises the backend they can only forge guest tokens, not real ones. Secrets are hashed, guest creation is rate limited (5/hour/IP). Only real accounts can call the merge endpoint so guests can't steal each other's data.

There are some downsides. If you clear your localStorage, you've lost access. It only works on one device. And I'll need some cleanup job eventually for abandoned guest accounts sitting in the DB.

I'd be interested in other's approaches to this. I wanted to make something that mirrored my real auth flow where everything starts from a valid refresh token (the real flow uses a cookie).

https://spikelog.com if you want to poke at it. Feel free to try and break it and please let me know if you do, or how I can tighten up my security.

Comments URL: https://news.ycombinator.com/item?id=46244742

Points: 4

# Comments: 0

1) Good point, I wanted to avoid the complexity of this for the first version, but you're 100% right, it would be great for someone to try first then upgrade when they register an account.

2) Thanks. Great to hear!

3) The best way to do this right now would be to create a project for each env and then give each env the API key from the corresponding project. Another way could be to put the env in the tags, but I think that's a bit messier as both lines would appear on the same chart (plus I've not even tested that works yet).

vibescaffold.dev looks interesting. Let me spend a bit of time and I'll feed back in your discord.

I’m very aware there are a lot of mature solutions in this space. Almost too much choice - which gave me decision paralysis when my need was simple. I’m not aiming to compete with full observability stacks like that. My goal with this was to intentionally stay on the extremely simple end of the spectrum. I'm aiming for something that’s quick to integrate, easy to understand, and focused on lightweight metrics rather than deep operational telemetry.

I’m also a big fan of Plausible and the idea of making select projects or charts publicly viewable in a dynamic way. Their public dashboard here was a big inspiration: https://plausible.io/plausible.io

That’s the model I borrowed for Spikelog as well: https://spikelog.com/p/spikelog

I also have this method for helping Cursor see the realtime output of services I'm running while developing locally which really speeds things up: https://foundinglean.substack.com/p/the-best-improvement-ive...

This substack is pretty new for me, but I'm planning on sharing more things there which may (or may not) help others. Next article there will just be sharing the symlinking setup (not rocket science - but some people don't know ln -s exists).

This really helps if they've had large version changes and the coding agents (without a search) only know about the old docs.

In that sense, adding prompt instructions for Spikelog feels like a natural extension of this trend. I mainly added them to share what worked for me when integrating another product with Spikelog.