rubenflamshep

I also tried Herdr. But I'm already using Aerospace, which makes MacOS window management a breeze and I don't need the remote session management stuff so the benefit wasn't quite there for me. It is a very thoughtful project so I did borrow lots of its features (Agents working now get an animation in my sketchy bar, a soft chime to indicate a finishes status).

My workflow has become a sort of "living process". When I encounter friction in some part of it, I focus my attention on it, it adapts, grows from from an annoyance to something that I delight in when using.

Some small examples: Google Chrome doesn't have default shortcuts for accessing bookmarks so I made an extension for myself that does this. I have a SketchyBar setup that I'm constantly updating. Recently added my CC/Codex plan usage to it so I'm no longer constantly running /status in the different CLIs. Etc, etc.

It's like that all that knowledge exists as a vague memory. I've found if I ask them specifics about any book I've read (w/o allowing tool calling), when they get it wrong their hallucinating things that are often correct or wrong in subtle ways. Are they're just completely out of pocket.

The issue remains the context window. I haven't seem any example of a "memories" system that the agent can update over time without it devolving into slop. The magic of the human brain is that we seem to have decently sophisticated heuristics parsing/saving memories as well as for letting memories that aren't accessed decay.

Looks cool! I think we're all vibe-coding our own menu bar usage apps :)

When I was going through this it was because OpenAI had defaulted to /fast mode with 2x token usage

Nah, if a human can't be bothered to write it themselves, I can't be bothered to read it.

Hmmm, I disagree. The AI is exceedingly average at everything it does and requires an expert human-in-the-loop to catch things that appear plausible but are slop. To that degree I think there is a difference between "AI generated" and slop.

Some engineers I work with have had less than desirable results with /simplify but it overall seems to work! I used to use some of the humanlayer subagents but they haven't been updated in several months

I find all of these .md files released by the labs to be ai generated slop. The only exception being maybe the /simplify command

This loads some bars and then just breaks for me :(

Very cool! One small UI grip: When I play a song the control at appears in the bottom of the page blocks the footer on the main page.

I assume you looked into automating the charts for new countries/years. What were the blockers for that?

There is no chart or table in the original paper. Feels like the one in the LLM-generated page is probably hallucinated?

I've found VSCode _ok_ to work with across across different workspaces/projects. The window memory is hit and miss. There's a secondary side bar I've been trying to NOT have open on startup but always seem to stick around. I'd prefer to programmatically manage the windows so I can tinker with an automated setup but the VSCode API/Plugins for managing this are terrible and tend to fail silently.

CLI within VSCode is workable but most of my VSCode envs are within a docker container. This is a pattern that I'm moving more and more away from as agents within a container kind of suck.

People are getting caught up in the "fast (but slow) diffusion)" that Dario has spoken to. Adoption of these tools has been fast but not instant but people will poke holes via "well, it hasn't done x yet".

For my own work I've focused on using the agents to help clean up our CICD and make it more robust, specifically because the rest of the company is using agents more broadly. Seems like a way to leverage the technology in a non-slop oriented way

Interesting. What would you say is your ratio of "sit down and make the implementation" time to "multi-agent system builds the thing" time?

Same. The only situation when I've consistently gotten a system to run for 20+ minutes was a data-analysis with tight guardrails and explicit multi-phase operations.

Outside that I'm juggling 2-3 sessions at most with nothing staying unattended for more than 10 minutes.

Quick plus one for Capital One after also working there. They're by far the most tech-forward of all the larger financial institutions, and by virtue of being a FI they take data-security much more seriously than any other "tech" companies.

No this is not a paid post lol

I'm on desktop!

I think the key here is having the network with the high-roller clients. The ability to execute is down funnel of that.

I wrote a local react app that lets you assign specific unit unit sounds from StarCraft II to different CC hooks: https://github.com/rubenflamshepherd/starcraft-claude

It also lets you manage Claude notifications more gracefully than what you get out of the box with CC. Been lazy about putting the finishing touches on it so this is a good kick in the ass to get that done!

It didn’t read as AI to me :)

As someone who has leaned fully into AI tooling this resonates. The current environment is an oligopoly so I'm learning how to leverage someone else's tool. However, in this way, I don't think LLMs are a radical departure from any proprietary other tool (e.g. Photoshop).

Hey, it's really nice you're supporting your wife like this! But there's nothing about how you're using AI as the "Lore Auditor" in the actual post?

1. I would allow users to play around without entering their email. 2. Why can't I just ask ChatGPT to do deep research on a topic for me? 3. What evals do you have to prove that the agents don't hallucinate output that sounds good

Your "How it works button" doesn't work friend

> That said, the main issue I find with agentic is my mental model getting desynchronized. No matter how fast the models get, it takes a fixed amount of time for me to catch up and understand what they've done.

This is why I'm so skeptical of anyone running 6+ Claude sessions at a time. I've gotten to 5 but really that was across 3 sessions with 2 standing by just to commit stuff. And even with just 3 sessions I constantly lost where I was and wasted time re-orienting myself, doing work in the wrong session, etc.

>The most enjoyable way I've found of staying synced is to stay in the driver's seat, and to command many small rapid edits manually.

Same, there's a fantastic flow state/momentum I can get in a single session just knocking off features. I don't mind switching between two sessions in this state but the experience is better when it's two different projects vs two different features on the same project. The complete context switch lets be re-orient more easily

They used to? I have a distinct memory of it doing exactly that a few months ago. Maybe it got dropped in the mad dash that passes for CC sprint cycles

> Without work people devolve into playing video games and smoking pot in their mom's basement.

I have no problem finding fulfilling and meaningful projects outside of my work! There are many people like me :)

Pretty good article until the bizarre post-script where they fall back on the tired "people derive meaning from their work" for why UBI is bad.