jasonriddle

It depends on what you are looking to get out of the next platform.

For me, I'm not interested in the social aspect of coding anymore, so I have a Synology NAS running a git server accessible via ssh and I push my code there.

I use klaus (https://github.com/jonashaag/klaus) as a read only git web ui. My NAS is connected to my tailscale network so it's easy to view things on the go. It's a simple setup and works great.

What coding harness are you using? What are some example workflows you have used either for? Have you used them only for new/simple projects or for more complicated refactoring or architecture design?

Thank you for the honest answer.

Yes, this is the conclusion I've come to as well. I don't want to continue supporting OpenAI nor Anthropic, but the other models don't seem to be anywhere close yet, despite the hype.

Slightly off topic, but question for folks.

I'm hoping to replace coding with Claude Sonnet 4.5 with a model with an open source or open weights model. Are any of the models on Ollama.com cloud offering (https://ollama.com/search?c=cloud) or any of the models on OpenRouter.ai a close replacement? I know that no model right now matches the full performance and capabilities of Claude Sonnet 4.5, but I want to know how close I can get and with which model(s).

If there is a model you say can replace it, talk about how long you have been using it for, and using what harness (Claude code, opencode, etc), and some strengths and weakness you have noticed. I'm not interested in what benchmarks say, I want to hear about real world use from programmers using these models.

Input:

(I copied and pasted text from "An Update on Heroku" https://www.heroku.com/blog/an-update-on-heroku/)

Output:

The Heroku Graveyard

Last Updated: February 06, 2026

Today, we’re officially putting Heroku on life support. We’re done building anything new and we’re just going to keep the lights on so it doesn't crash or get hacked. We’re calling it "operational excellence" because that sounds better than "we've given up on this product." I know this sucks, so let me try to spin it for you.

If you’re already paying us with a credit card, don't worry—we’ll keep taking your money. Everything will stay exactly the same because we aren't touching a single line of code anymore. You can keep running your business on this ghost ship if you want.

If you’re a big company, forget it. We aren't signing any new enterprise deals. If you already have one, we’ll take your renewals until you finally get fed up and migrate somewhere else.

Why we’re doing this We’re gutting the budget for this and moving all the engineers to the AI hype train because that’s where the easy money is right now.

That makes sense.

I think I tried using it maybe 4 years ago or so, and I had the same feeling. It just felt.. awkward to use, lots of friction. I was hoping it had changed by now, but I guess that hasn't happened.

Ah that's disappointing.

So you've just been with 1Password then? Did you try KeePass or anything else?

It's been little things and mainly usability/polish things.

Sometimes the vault doesn't unlock and I have to enter in my password 2-3 times.

It doesn't always capture all information from a page properly when creating a new login and there are additional fields to capture.

The "detecting if a website supports key passes and one time password" feature for Watchtower was overwhelming with lots of information, until I clicked each one and had to ignore it.

These reasons alone are not enough for me to leave, the 3 big problems are below.

1 - I was feeling more uncomfortable having websites promote using passkeys, and I would store that in 1Password, but then I wasn't sure if 1Password as going to make it easy to migrate that stuff out. So, I want to use something open source, so I don't have to worry about losing access/managing that stuff in a propertiery/closed product. It might be easy to export/migrate out today, until something changes and they no longer allow that or make it very difficult/hard to scale/automate.

2 - I have a strong feeling this price increase is being justified by "AI" somehow. I'm sure, like all other companies, 1Password is internally forcing/requiring its developers to use coding models, and sonnet, opus, etc are expensive to use and the cost adds up. Also, I don't like the direction of where things are headed, where people are becoming more relaxed and not reviewing code properly and merging in code that will cause security issues later (perhaps openclaw fits into this bucket) or they are taking open-source code they laundering it for companies internally to use (I can't prove this, but if a model is trained on public data/code, it seems very likely). Something about that just bothers me especially when a company is worth billions of dollars.

3 - I've spent the last 3 years building up my homelab and using Pikapods for hosting various things. I want to support open-source more and run my own things and pay supporters properly to maintain things. I've always been a bit nervous what might happen if 1Password gets hacked, either because of poor security or due to a third party vendor. I still have the problem of my things getting hacked, but I pay more attention to how I secure things and use Tailscale and not publish things on the broad internet (when it makes sense). Also, I would be a hypocrite to dismiss the value of coding llms, as I'm using them myself. But how I'm using them, I'm using them to do security reviews of my docker compose files or kubernetes yaml files. Having coding llms has made it so much easier to maintain a homelab.

Yeah, pretty disappointed by this as well. The app has been getting buggier overtime and I was already considering leaving, so this was the push I needed.

Seems like the most popular players in this space are Bitwarden and KeePass, does anybody have a positive or negative experience to share with either?

Thanks for making this. Is the code for this available somewhere public?

What AI stuff are you referring to? I just learned about this project from this blog post, so I don't have the full context on their AI work.

This looks interesting.

Instead of having to code this up using typescript, is there an MCP server or API endpoint I can use?

Basically, I want to connect an MCP server to an agent, tell it it can run typescript code in order to solve a problem or verify something.

I've been unemployed for about 1 year now. I was in SF working in tech for about 7 years, and decided I don't want to do that anymore, so I quit.

It's been tough. The hardest part about being unemployed is it is very hard to structure your days because work is no longer the thing that is forcing you to get up, get out, go to bed on time, etc. It's also a strange feeling having to spend from your savings/emergency fund without money coming in, you feel bad and guilty for doing so, it's weird.

I'm changing careers. I've always liked teaching, so I'm doing volunteer english teaching while preparing to apply to go back to school in order to get a Masters in Education.

In the mean time, I'm also doing other small things. Learning about AI, going to board game meetups, doing some traveling, overall it's not the most fun part of my life, but I'm treating it as I will look back on this and realize this was necessary.

Interesting.

Here is a feature request. I want a Google chrome extension so that as I'm browsing the web, similar to how pocket used to work, I can bookmark a page to read more about it later.

So, it looks like this might work with fly.io?

fly.io provides a way to connect to their servers via wireguard (https://fly.io/docs/blueprints/connect-private-network-wireg...), and so tailguard could connect to their wireguard instance?

Hi, thanks for sharing.

My main concern with these browser agents are how are they handling prompt injection. This blog post on Perplexity's Comet browser comes to mind: https://brave.com/blog/comet-prompt-injection/.

Also, today Anthropic announced Claude for Chrome (https://www.anthropic.com/news/claude-for-chrome) and from the discussion on that (https://news.ycombinator.com/item?id=45030760), folks quickly pointed out that the attack success rate was 11.2%, which still seems very high.

How do you plan to handle prompt injection?

I'm using linkding: https://linkding.link/.

It works great. It has a minimal set of features and can be self-hosted.

I'm paying pikapods to host it for me, but if I needed to, I can switch to doing it on my own.

Sure, but to clarify, so you are probably setting temperature to close to 0 in order to try to get as consistent output as possible based on the input? Have you made any changes to top k and/or top p that you have found makes agents output more consistent/deterministic?

When you say "same output" in

Every subagent call should be like calling a pure function. Same input, same output. No shared memory. No conversation history. No state.

How are you setting temperature, top k, top p, etc?

Nice! I have been searching for something like this for while!

This is a great post, thank you for sharing. I like the idea of giving hints to the LLMs.

To clarify, the example that was provided using `command_not_found_handler`, is that possible to implement in bash? Or perhaps you were saying this would be a nice to have if this functionality existed?

@simonw, how much have you used Claude code? I'm wondering if it makes sense to use Claude code as the main driver, but have crafted specialized "llm" based tools for Claude to use.

I would love to only have to use llm, but with the amount of functionality that Claude code provides, it seems unlikely that llm would ever support all of its features. Also, I have realized that llm is narrower in scoop, which is something I like about the tool.

Yes,

https://www.intuit.com/privacy/statement/

> How we share your personal information

For mergers and acquisitions. If we are involved with a merger, asset sale, financing, liquidation, bankruptcy, or the acquisition of all or part of our business to another company, we may share your personal information with that company and its advisors before and after the transaction date.

But unlike every other company, this company has your most sensitive tax information. For some that's fine, but others might not like that.

Just so that you are aware, TaxHawk (which owns and operates FreeTaxUSA) may choose to sell your information in the event of a "business transition" (bankruptcy, merger, etc)

From https://www.freetaxusa.com/privacy

> Business transitions

In the event TaxHawk evaluates or conducts a business transition whether as a going concern or as part of bankruptcy, liquidation, or similar proceeding, such as a merger, being acquired by another company, or selling a portion of its assets, the personal information of users will, in most instances, be part of the assets transferred. Users will be notified, via email and/or through a prominent notice on the FreeTaxUSA.com website, 30 days prior to a change of ownership or control of their personal information.

To add onto 3, I do my taxes every year using either the desktop version of Turbo Tax or H&R Block. They only make Mac or Windows versions of their software.

I've read online that people have attempted to use wine in order to emulate the Windows environment with no success.

I've been looking to migrate away from https://simplenote.com to something else and I've been dragging my feet. However, what's happened in the past 2 weeks with the CEO was the push I needed to resume looking for another solution.

I've landed on Obsidian.md. I actually prefer Simplenote over Obisidian, because Simplenote feels smoother and the online web app is great, but I've decided that having more control over my own data is more important so Obsidian it is.

I'm of the mindset that NixOS is too much, all at once. I understand the power of the Nix ecosystem, but the barrier of entry remains too high to enter.

I've switched to using devenv[1] as a first step. This at least allows me to experiment with using nix to configure my dev environment in a per directory fashion, while allowing me to use the Linux distro of my choice.

[1] https://github.com/cachix/devenv

I've been using Numberbarn.

I've been liking it so far, no issues to report. However, I've only been using it for two months now.

Also, similar to Google Voice, some businesses won't let you use a VOIP number for your phone number.

There was an article someone posted on hacker news previously (I couldn't find the link unfortunately, maybe someone can help me out).

It didn't refer to developed nations, but it was saying that there are companies that were already planting lots of trees, but the type of the trees they were planting were negatively impacting the biodiversity.

I don't remember if that meant that the trees weren't native to that region which was leading to problems or if they only picked two types of trees to plant (perhaps because they were the cheapest) which may not be diverse enough.