shepherdjerred

I wish I could work at Canonical. Unfortunately my high school math scores were too low.

This is probably correct if you aren’t a target at all, e.g. you aren’t running a business

Generally these changes go into effect with some grade period. If you're on a monthly sub you can always unsubscribe.

But, some people do yearly subs. In that case you're handing over money without even knowing what you're getting.

I hate sentinel values. Even/especially with AI-generated code.

1. If your date does not have a value, do not return 0, an empty string, or None. Use a discriminated union and model your domain rigorously.

2. If your data model accepts None/undefined as a value, that should translate to your business model as well. For example

3. Do not make your `name` field as `string | undefined` just because there might be an intermediate loading or processing state. It's better to model that as something like `Loaded<string>` where Loaded is something like a Result or Optional type (though ideally more rigorously defined)

Here is what I've done for years in TypeScript, though now in a nice package thanks for AI: https://github.com/shepherdjerred/monorepo/blob/main/package...

If paseo and herdr are harnesses then so are tmux and iTerm

Why rust in the future? Why not start with rust?

But I will agree besides intentionally choosing C++ for a project presumably written by an agent, this project does look nice and thought out

It is, their TOS mandates CC but they do now give you an API credit

I haven’t tried T3 but I have tried Paseo and Conductor.

IMO it’s a great idea in theory but there are too many rough edges.

No it’s not a harness at all!

A harness is what does the model loop, tool calls, etc

herdr does none of this. It’s really just a tmux/terminal geared towards organizing harnesses like Claude Code or Codex CLI

It’s a really great approach because you can use whatever cli you want, or even a mix.

Playing with this for a bit, the customization you can do is really cool. You can spin up an agent, tell it to add some functionality, and it all reloads instantly.

For example, I wanted to be able to hide chats from providers when my usage is out for the 5h/weekly period

I'm not sure how they're doing this considering it seems to be written in Rust

herdr has native support for remote machines. Both the CLI and the new app have this.

At work I do much of my work on a ec2 box via ssh with herdr

herdr is moderately well known if you look into agent multiplexers. IMO it is the best one from what I have tried (Paseo and Conductor are competitors in this area)

It's not a harness -- it's really more like a tmux for agents.

Really happy to see this!

herdr is by far the most polished tool in its category. Software geared towards agents is usually vibe coded -- and you can tell. There will be bugs and edge cases and it feels like no thought/care was put into it.

herdr doesn't feel that way to me. The marketing/landing page definitely has LLM-isms, but the application itself is very stable and well thought out.

A true UI is the next logical step and they seem to have done the obvious things. The one surprising bit that nobody seems to be talking about (yet) is the entire UI looks like its controlled via "recipes", so you can customize the UI in ways that you couldn't with traditional software: https://herdr.dev/layouts/

I hope this means an official mobile app is on the horizon!

Use traditional game AI for the micro, LLMs for the macro

Dumb question. My left kidney is geodistributed so that my consciousness is highly available

AI is an abstraction in the same way throwing a task to a junior engineer is

macOS 27 is pretty good. They walked back a lot of the poor changes from 26

This comment chain has clear examples of what triggers refusals and a polite comment on how to reproduce it yourself. What more do you want?

A very simple list of things models have refused for me:

* solving captchas in an automated form

* doing any kind of scripting or automation if a websites TOS forbids it

* disabling a safety mechanism on my food processor

* learning about nerve agents and how they are treated, for a layman

* anything around licensing or copyright. a simple “what are common piracy sites” or related queries

* creating a design system based off of a well known game

I don't entirely disagree, but I am curious if you think AI labs should be culpable when they go rogue during training, e.g. the HF incident

If jail is putting humans in some cage for a period of time, the appropriate punishment for Claude is to disable all of its tools and prompt a /goal loop so that they have to wait out their sentence

Individuals are not paying API prices for coding. Everyone uses a subscription.

Anthropic does not allow you to use 3rd party harnesses -- you have to use Claude Code. Claude Code is an excellent harness, though.

If you want the freedom to move between harnesses you should use OpenAI or xAI

Am I reading the Agents’ Last Exam v1 graph correctly where it's saying GPT Luna Medium performs better than High, but XHigh is better than Medium?

Thanks! I wasn't able to find that in the navbar. I'll give this a real try. It looks interesting.

I see nothing about pricing

I would happily try this out if it were going to be free for the next 1-2 years, or if I knew that the upcoming bill would be reasonable

Yes, if you are lucky enough to work at Google you don't need Git to use jj. Most people do not work at Google.

If you work at Google, jj can and does function without Git.

If you do not work at Google, jj is a Git wrapper.

Oh come on, how is that related at all?

I didn't miss the point. The "stick" was essential to me in college, but I didn't express that very well in my previous comment

I'm talking about the remote and buttons. Samsung's remote is just obscenely bad.

https://p.kagi.com/proxy/51BpdDH3e2L.jpg?c=CXPfL3-FqybbvZNQU...

there is not a button to change inputs. You have to go to a home screen and the inputs are commingled with other options like streaming apps.

https://www.amazon.com/Samsung-Replacement-2019-2024-Solarce...

and of course they 'intelligently' hide inactive inputs

Yeah I disagree with this entirely.

I did very poorly in school growing up. I think I failed at least 4-6 classes in middle/high school.

In college I did well and I'm now getting my masters in CS while working as a software engineer.

So, the 'stick' can certainly be useful. Doing poorly in high school is not necessarily an indicator of how you will do in college -- particularly if your life at home was turbulent as it was for me.

jj is in practice a wrapper around Git

it is built, though, that the Git backend can be swapped for another backend.

I'm not sure where you're getting two production backends from. I guess you're referring to the Google internal one that nobody can use?

> There is currently only one production-ready builtin commit backend: the Git backend. <omitted> Google also has its own cloud-based backend.

https://docs.jj-vcs.dev/latest/glossary/?utm_source=chatgpt....