mcapodici

Docker question. Who prefers docker for local dev vs. run the thing directly. I usually prefer run the thing for debugging and hot reloading. Sure: build the containers and test them in CI. Dependencies like PG sure run in containers as they change less often and you save the install.

Working code that people pay to use is of course an asset. If it is a liability then just delete it all an have no source code an you just paid off a debt for free.

I wonder if it is almost like writing. In that to make the diagram that communicates well you need to have solved the problem properly.

They outsmarted you!

Sounds cheaper than AWS :)

Most of my connection issues are between the street and my computer, I don't think Cloudflare got there. Yet.

Yes and then the choice is run it and forgive it, or, stop the process midway.

If you stop then you have to decide whether to charge for uncompleted work.

Interesting tradeoffs.

For very small ops e.g. individual Lambda invocation you have similar concerns especially if lots are fired at once from a queue or schedule or fanout.

I don't use memory. I am really not keen on the idea of having this hidden context fed into the LLM, and I prefer to use docs, both markdown and stored in a wiki like Confluence. Using repos and wikis also lets you scope the information so hyper-specific memory doesn't affect other projects.

If I want the LLM to remember something I ask it to update some docs, and even check that docs are consistent across the board after doing so.

The only thing I want the LLM to remember everywhere is talk like a human (no load-bearing, not this/that etc...), so I have an AGENT.md for that.

Use instead:

   npx skills add mattpocock/skills --skill=grill-me

or

   npx skills add imbue-ai/blueprint

Or anything similar.

1052 pts on HN. I'd say they are pretty good at marketing actually.

While they may not initially done well they are certainly riding this wave.

For blogging I just don't use an LLM at all. I just use a spelling/grammar checker. I get to sound like me even if that is imperfect.

For docs I have used an LLM for the convenience of having them kept up to date.

Ok for example my strategy is buy Moderna. That's it. Backtest that over the last year.

Agreed on 2nd point. Especially with no real hypothesis i.e. just throw AI at it.

The problem is the bots may have some of the backtest data already in their training.

Unless you are very short on space, keep all 10. Because 50-100% of them will randomly cease to work.

I always forget the US doesn't use A4 etc. (or do you?)

What a time to be alive stuff.

https://github.com/anthropics/fermats-last-theorem/blob/main...

It is worse than even that:

> Within the space of four seconds, the energy from the annulus gas flow accelerated the IP turbine past its critical speed, until centrifugal forces exceeded the ultimate strength of the nickel alloy disk. The red-hot, wildly spinning disk instantly fractured into several sections, which rocketed outward in multiple directions at incomprehensible speed.

cat is only for concatenation so it won't combine e.g. XML files, images, zip files etc. unless those files can just be splatted together.

Little did he know in 2026 you can feed that cartoon into AI and it will build you that app

I haven't checked the 12 factors for a while but feel proud to have worked on systems probably for a long old while where most or all have applied. To the point I'd naturally do these things without consciously thinking.

I think that says a lot about good practices spreading than anything else. To not do these things: maybe a startup moving fast, or a very isolated company or just some old legacy COBOL type thing where you want the thing to still work as the main concern.

That has to be the most absurd but technically correct thing I have read.

TUIs allow you to have infinitely many running in tmux and also work over ssh so there is that.

Thanks! And absolutely there is a big tidy up and code review on the horizon for this repo.

Data storage changed direction a few times as I heavily used this and found things that seemed fine in theory didn't work out in practice.

A big one was using syncthing and folder sharing in the browser had all sorts of issues for this app and how it operates.

I'll probably add a sign soon to say this is alpha (as in before beta, a prototype) software.

It is vibe coded and is also an experiment is using AI more wildly than I would at work

Thanks for wording that kindly and thoughtfully. Some may have said it is trash and moved on :)

Hi. As you are deeply thinking about these sorts of things, I want to see what you think of my https://useorganizer.com which turns the todo idea on its head a bit by being timeline oriented.

That is an amazing idea. I can imagine this getting popular. It could be used for faster local dev loops and porting GHA to other servers. I imagine for data/compute residency, cost, performance, vm types etc.

A new tech YouTube channel. It might be all AI or not but we'll see!

https://m.youtube.com/@MartinCapodici

Hi msaltz, I like your other introspection posts alot.

All these words are too broad is why. It allows someone to say something without really knowing what they mean. But also they may know what they mean, so you are not sure!

Device is a handy one though you often want to say "computer, laptop, tablet or phone" in one go. We could call them all computers but that would confuse people I think.

Yes I agree. Content is an empty word, and can mean anything, and sort of means nothing. I get why large corporations would want to use the term as they do everything in bulk and need a general term, but for a solo writer, youtuber, etc, better to think of it as the thing, e.g. "tutorial", "podcast", "exploration".

I was experimenting and found deepseek-v4-flash and found it cheap (way cheap compared to sota models) and perfectly good at tool calling. I did a post on it https://martincapodici.com/2026/07/18/weekly-ai-learnings-3/

So $1.5 for 40m tokens I guess would cost much more with sota (but would need less tokens perhaps).